Daily Papers

We introduce a novel method for pre-training of large-scale vision encoders. Building on recent advancements in autoregressive pre-training of vision models, we extend this framework to a multimodal setting, i.e., images and text. In this paper, we present AIMV2, a family of generalist vision encoders characterized by a straightforward pre-training process, scalability, and remarkable performance across a range of downstream tasks. This is achieved by pairing the vision encoder with a multimodal decoder that autoregressively generates raw image patches and text tokens. Our encoders excel not only in multimodal evaluations but also in vision benchmarks such as localization, grounding, and classification. Notably, our AIMV2-3B encoder achieves 89.5% accuracy on ImageNet-1k with a frozen trunk. Furthermore, AIMV2 consistently outperforms state-of-the-art contrastive models (e.g., CLIP, SigLIP) in multimodal image understanding across diverse settings.

OpenAI's CLIP, released in early 2021, have long been the go-to choice of vision encoder for building multimodal foundation models. Although recent alternatives such as SigLIP have begun to challenge this status quo, to our knowledge none are fully open: their training data remains proprietary and/or their training recipes are not released. This paper fills this gap with OpenVision, a fully-open, cost-effective family of vision encoders that match or surpass the performance of OpenAI's CLIP when integrated into multimodal frameworks like LLaVA. OpenVision builds on existing works -- e.g., CLIPS for training framework and Recap-DataComp-1B for training data -- while revealing multiple key insights in enhancing encoder quality and showcasing practical benefits in advancing multimodal models. By releasing vision encoders spanning from 5.9M to 632.1M parameters, OpenVision offers practitioners a flexible trade-off between capacity and efficiency in building multimodal models: larger models deliver enhanced multimodal performance, while smaller versions enable lightweight, edge-ready multimodal deployments.

In recent years, notable advancements have been made in the domain of visual document understanding, with the prevailing architecture comprising a cascade of vision and language models. The text component can either be extracted explicitly with the use of external OCR models in OCR-based approaches, or alternatively, the vision model can be endowed with reading capabilities in OCR-free approaches. Typically, the queries to the model are input exclusively to the language component, necessitating the visual features to encompass the entire document. In this paper, we present VisFocus, an OCR-free method designed to better exploit the vision encoder's capacity by coupling it directly with the language prompt. To do so, we replace the down-sampling layers with layers that receive the input prompt and allow highlighting relevant parts of the document, while disregarding others. We pair the architecture enhancements with a novel pre-training task, using language masking on a snippet of the document text fed to the visual encoder in place of the prompt, to empower the model with focusing capabilities. Consequently, VisFocus learns to allocate its attention to text patches pertinent to the provided prompt. Our experiments demonstrate that this prompt-guided visual encoding approach significantly improves performance, achieving state-of-the-art results on various benchmarks.

Humans possess the remarkable skill of Visual Perception, the ability to see and understand the seen, helping them make sense of the visual world and, in turn, reason. Multimodal Large Language Models (MLLM) have recently achieved impressive performance on vision-language tasks ranging from visual question-answering and image captioning to visual reasoning and image generation. However, when prompted to identify or count (perceive) the entities in a given image, existing MLLM systems fail. Working towards developing an accurate MLLM system for perception and reasoning, we propose using Versatile vision enCoders (VCoder) as perception eyes for Multimodal LLMs. We feed the VCoder with perception modalities such as segmentation or depth maps, improving the MLLM's perception abilities. Secondly, we leverage the images from COCO and outputs from off-the-shelf vision perception models to create our COCO Segmentation Text (COST) dataset for training and evaluating MLLMs on the object perception task. Thirdly, we introduce metrics to assess the object perception abilities in MLLMs on our COST dataset. Lastly, we provide extensive experimental evidence proving the VCoder's improved object-level perception skills over existing Multimodal LLMs, including GPT-4V. We open-source our dataset, code, and models to promote research. We open-source our code at https://github.com/SHI-Labs/VCoder

The deployment of foundation models for medical imaging has demonstrated considerable success. However, their training overheads associated with downstream tasks remain substantial due to the size of the image encoders employed, and the inference complexity is also significantly high. Although lightweight variants have been obtained for these foundation models, their performance is constrained by their limited model capacity and suboptimal training strategies. In order to achieve an improved tradeoff between complexity and performance, we propose a new framework to improve the performance of low complexity models via knowledge distillation from multiple large medical foundation models (e.g., MedSAM, RAD-DINO, MedCLIP), each specializing in different vision tasks, with the goal to effectively bridge the performance gap for medical image segmentation tasks. The agglomerated model demonstrates superior generalization across 12 segmentation tasks, whereas specialized models require explicit training for each task. Our approach achieved an average performance gain of 2\% in Dice coefficient compared to simple distillation.

Multimodal language models (MLMs) integrate visual and textual information by coupling a vision encoder with a large language model through the specific adapter. While existing approaches commonly rely on a single pre-trained vision encoder, there is a great variability of specialized encoders that can boost model's performance in distinct domains. In this work, we propose MOVE (Mixture of Vision Encoders) a simple yet effective approach to leverage multiple pre-trained encoders for specialized multimodal tasks. MOVE automatically routes inputs to the most appropriate encoder among candidates such as Unichat, InternViT, and Texify, thereby enhancing performance across a diverse set of benchmarks, including ChartQA, MMBench, and MMMU. Experimental results demonstrate that MOVE achieves competitive accuracy without incurring the complexities of image slicing for high-resolution images.

Enhanced visual understanding serves as a cornerstone for multimodal large language models (MLLMs). Recent hybrid MLLMs incorporate a mixture of vision experts to address the limitations of using a single vision encoder and excessively long visual tokens. Despite the progress of these MLLMs, a research gap remains in effectively integrating diverse vision encoders. This work explores fusion strategies of visual tokens for hybrid MLLMs, leading to the design of LEO, a novel MLLM with a dual-branch vision encoder framework that incorporates a post-adaptation fusion strategy and adaptive tiling: for each segmented tile of the input images, LEO sequentially interleaves the visual tokens from its two vision encoders. Extensive evaluation across 13 vision-language benchmarks reveals that LEO outperforms state-of-the-art open-source MLLMs and hybrid MLLMs on the majority of tasks. Furthermore, we show that LEO can be adapted to the specialized domain of autonomous driving without altering the model architecture or training recipe, achieving competitive performance compared to existing baselines. The code and model will be publicly available.

Our approach, which we call Embeddings for Language/Image-aligned X-Rays, or ELIXR, leverages a language-aligned image encoder combined or grafted onto a fixed LLM, PaLM 2, to perform a broad range of tasks. We train this lightweight adapter architecture using images paired with corresponding free-text radiology reports from the MIMIC-CXR dataset. ELIXR achieved state-of-the-art performance on zero-shot chest X-ray (CXR) classification (mean AUC of 0.850 across 13 findings), data-efficient CXR classification (mean AUCs of 0.893 and 0.898 across five findings (atelectasis, cardiomegaly, consolidation, pleural effusion, and pulmonary edema) for 1% (~2,200 images) and 10% (~22,000 images) training data), and semantic search (0.76 normalized discounted cumulative gain (NDCG) across nineteen queries, including perfect retrieval on twelve of them). Compared to existing data-efficient methods including supervised contrastive learning (SupCon), ELIXR required two orders of magnitude less data to reach similar performance. ELIXR also showed promise on CXR vision-language tasks, demonstrating overall accuracies of 58.7% and 62.5% on visual question answering and report quality assurance tasks, respectively. These results suggest that ELIXR is a robust and versatile approach to CXR AI.

Vision language models (VLMs) typically pair a modestly sized vision encoder with a large language model (LLM), e.g., Llama-70B, making the decoder the primary computational burden during training. To reduce costs, a potential promising strategy is to first train the vision encoder using a small language model before transferring it to the large one. We construct small "surrogate models" that share the same embedding space and representation language as the large target LLM by directly inheriting its shallow layers. Vision encoders trained on the surrogate can then be directly transferred to the larger model, a process we call zero-shot grafting -- when plugged directly into the full-size target LLM, the grafted pair surpasses the encoder-surrogate pair and, on some benchmarks, even performs on par with full decoder training with the target LLM. Furthermore, our surrogate training approach reduces overall VLM training costs by ~45% when using Llama-70B as the decoder.

The ability to accurately interpret complex visual information is a crucial topic of multimodal large language models (MLLMs). Recent work indicates that enhanced visual perception significantly reduces hallucinations and improves performance on resolution-sensitive tasks, such as optical character recognition and document analysis. A number of recent MLLMs achieve this goal using a mixture of vision encoders. Despite their success, there is a lack of systematic comparisons and detailed ablation studies addressing critical aspects, such as expert selection and the integration of multiple vision experts. This study provides an extensive exploration of the design space for MLLMs using a mixture of vision encoders and resolutions. Our findings reveal several underlying principles common to various existing strategies, leading to a streamlined yet effective design approach. We discover that simply concatenating visual tokens from a set of complementary vision encoders is as effective as more complex mixing architectures or strategies. We additionally introduce Pre-Alignment to bridge the gap between vision-focused encoders and language tokens, enhancing model coherence. The resulting family of MLLMs, Eagle, surpasses other leading open-source models on major MLLM benchmarks. Models and code: https://github.com/NVlabs/Eagle

Vision Language Models (VLMs) have achieved remarkable progress in multimodal tasks, yet they often struggle with visual arithmetic, seemingly simple capabilities like object counting or length comparison, which are essential for relevant complex tasks like chart understanding and geometric reasoning. In this work, we first investigate the root causes of this deficiency through a suite of probing tasks focusing on basic visual arithmetic. Our analysis reveals that while pre-trained vision encoders typically capture sufficient information, the text decoder often fails to decode it correctly for arithmetic reasoning. To address this, we propose CogAlign, a novel post-training strategy inspired by Piaget's theory of cognitive development. CogAlign trains VLMs to recognize invariant properties under visual transformations. We demonstrate that this approach significantly improves the performance of three diverse VLMs on our proposed probing tasks. Furthermore, CogAlign enhances performance by an average of 4.6% on CHOCOLATE and 2.9% on MATH-VISION, outperforming or matching supervised fine-tuning methods while requiring only 60% less training data. These results highlight the effectiveness and generalizability of CogAlign in improving fundamental visual arithmetic capabilities and their transfer to downstream tasks.

Multi-modal Large Language Models (MLLMs) have made significant strides in expanding the capabilities of Large Language Models (LLMs) through the incorporation of visual perception interfaces. Despite the emergence of exciting applications and the availability of diverse instruction tuning data, existing approaches often rely on CLIP or its variants as the visual branch, and merely extract features from the deep layers. However, these methods lack a comprehensive analysis of the visual encoders in MLLMs. In this paper, we conduct an extensive investigation into the effectiveness of different vision encoders within MLLMs. Our findings reveal that the shallow layer features of CLIP offer particular advantages for fine-grained tasks such as grounding and region understanding. Surprisingly, the vision-only model DINO, which is not pretrained with text-image alignment, demonstrates promising performance as a visual branch within MLLMs. By simply equipping it with an MLP layer for alignment, DINO surpasses CLIP in fine-grained related perception tasks. Building upon these observations, we propose a simple yet effective feature merging strategy, named COMM, that integrates CLIP and DINO with Multi-level features Merging, to enhance the visual capabilities of MLLMs. We evaluate COMM through comprehensive experiments on a wide range of benchmarks, including image captioning, visual question answering, visual grounding, and object hallucination. Experimental results demonstrate the superior performance of COMM compared to existing methods, showcasing its enhanced visual capabilities within MLLMs. Code will be made available at https://github.com/YuchenLiu98/COMM.

Despite significant advancements in Multimodal Large Language Models (MLLMs) for understanding complex human intentions through cross-modal interactions, capturing intricate image details remains challenging. Previous methods integrating multiple vision encoders to enhance visual detail introduce redundancy and computational overhead. We observe that most MLLMs utilize only the last-layer feature map of the vision encoder for visual representation, neglecting the rich fine-grained information in shallow feature maps. To address this issue, we propose \modelname, a simple yet effective multi-layer feature fuser that efficiently integrates deep and shallow features from Vision Transformers (ViTs). Specifically, it leverages semantically aligned deep features as queries to dynamically extract missing details from shallow features, thus preserving semantic alignment while enriching the representation with fine-grained information. Applied to the LLaVA-1.5 model, \modelname~achieves significant improvements in visual representation and benchmark performance, providing a more flexible and lightweight solution compared to multi-encoder ensemble methods. The code and model have been released at https://github.com/yuecao0119/MMFuser.

Large Vision-Language Models (LVLMs) rely on vision encoders and Large Language Models (LLMs) to exhibit remarkable capabilities on various multi-modal tasks in the joint space of vision and language. However, the Typographic Attack, which disrupts vision-language models (VLMs) such as Contrastive Language-Image Pretraining (CLIP), has also been expected to be a security threat to LVLMs. Firstly, we verify typographic attacks on current well-known commercial and open-source LVLMs and uncover the widespread existence of this threat. Secondly, to better assess this vulnerability, we propose the most comprehensive and largest-scale Typographic Dataset to date. The Typographic Dataset not only considers the evaluation of typographic attacks under various multi-modal tasks but also evaluates the effects of typographic attacks, influenced by texts generated with diverse factors. Based on the evaluation results, we investigate the causes why typographic attacks may impact VLMs and LVLMs, leading to three highly insightful discoveries. By the examination of our discoveries and experimental validation in the Typographic Dataset, we reduce the performance degradation from 42.07% to 13.90% when LVLMs confront typographic attacks.

Existing vision-language models (VLMs) mostly rely on vision encoders to extract visual features followed by large language models (LLMs) for visual-language tasks. However, the vision encoders set a strong inductive bias in abstracting visual representation, e.g., resolution, aspect ratio, and semantic priors, which could impede the flexibility and efficiency of the VLMs. Training pure VLMs that accept the seamless vision and language inputs, i.e., without vision encoders, remains challenging and rarely explored. Empirical observations reveal that direct training without encoders results in slow convergence and large performance gaps. In this work, we bridge the gap between encoder-based and encoder-free models, and present a simple yet effective training recipe towards pure VLMs. Specifically, we unveil the key aspects of training encoder-free VLMs efficiently via thorough experiments: (1) Bridging vision-language representation inside one unified decoder; (2) Enhancing visual recognition capability via extra supervision. With these strategies, we launch EVE, an encoder-free vision-language model that can be trained and forwarded efficiently. Notably, solely utilizing 35M publicly accessible data, EVE can impressively rival the encoder-based VLMs of similar capacities across multiple vision-language benchmarks. It significantly outperforms the counterpart Fuyu-8B with mysterious training procedures and undisclosed training data. We believe that EVE provides a transparent and efficient route for developing a pure decoder-only architecture across modalities. Our code and models are publicly available at: https://github.com/baaivision/EVE.

Large vision language models (VLMs) combine large language models with vision encoders, demonstrating promise across various tasks. However, they often underperform in task-specific applications due to domain gaps between pre-training and fine-tuning. We introduce VITask, a novel framework that enhances task-specific adaptability of VLMs by integrating task-specific models (TSMs). VITask employs three key strategies: exemplar prompting (EP), response distribution alignment (RDA), and contrastive response tuning (CRT) to improve the task-specific performance of VLMs by adjusting their response distributions. EP allows TSM features to guide VLMs, while RDA enables VLMs to adapt without TSMs during inference by learning from exemplar-prompted models. CRT further optimizes the ranking of correct image-response pairs, thereby reducing the risk of generating undesired responses. Experiments on 12 medical diagnosis datasets across 9 imaging modalities show that VITask outperforms both vanilla instruction-tuned VLMs and TSMs, showcasing its ability to integrate complementary features from both models effectively. Additionally, VITask offers practical advantages such as flexible TSM integration and robustness to incomplete instructions, making it a versatile and efficient solution for task-specific VLM tuning. Our code are available at https://github.com/baiyang4/VITask.

As the key component in multimodal large language models (MLLMs), the ability of the visual encoder greatly affects MLLM's understanding on diverse image content. Although some large-scale pretrained vision encoders such as vision encoders in CLIP and DINOv2 have brought promising performance, we found that there is still no single vision encoder that can dominate various image content understanding, e.g., the CLIP vision encoder leads to outstanding results on general image understanding but poor performance on document or chart content. To alleviate the bias of CLIP vision encoder, we first delve into the inherent behavior of different pre-trained vision encoders and then propose the MoVA, a powerful and novel MLLM, adaptively routing and fusing task-specific vision experts with a coarse-to-fine mechanism. In the coarse-grained stage, we design a context-aware expert routing strategy to dynamically select the most suitable vision experts according to the user instruction, input image, and expertise of vision experts. This benefits from the powerful model function understanding ability of the large language model (LLM) equipped with expert-routing low-rank adaptation (LoRA). In the fine-grained stage, we elaborately conduct the mixture-of-vision-expert adapter (MoV-Adapter) to extract and fuse task-specific knowledge from various experts. This coarse-to-fine paradigm effectively leverages representations from experts based on multimodal context and model expertise, further enhancing the generalization ability. We conduct extensive experiments to evaluate the effectiveness of the proposed approach. Without any bells and whistles, MoVA can achieve significant performance gains over current state-of-the-art methods in a wide range of challenging multimodal benchmarks. Codes and models will be available at https://github.com/TempleX98/MoVA.

The recent advent of Large Language Models (LLMs) has ushered sophisticated reasoning capabilities into the realm of video through Video Large Language Models (VideoLLMs). However, VideoLLMs currently rely on a single vision encoder for all of their visual processing, which limits the amount and type of visual information that can be conveyed to the LLM. Our method, MERV, Multi-Encoder Representation of Videos, instead leverages multiple frozen visual encoders to create a unified representation of a video, providing the VideoLLM with a comprehensive set of specialized visual knowledge. Spatio-temporally aligning the features from each encoder allows us to tackle a wider range of open-ended and multiple-choice video understanding questions and outperform prior state-of-the-art works. MERV is up to 3.7% better in accuracy than Video-LLaVA across the standard suite video understanding benchmarks, while also having a better Video-ChatGPT score. We also improve upon SeViLA, the previous best on zero-shot Perception Test accuracy, by 2.2%. MERV introduces minimal extra parameters and trains faster than equivalent single-encoder methods while parallelizing the visual processing. Finally, we provide qualitative evidence that MERV successfully captures domain knowledge from each of its encoders. Our results offer promising directions in utilizing multiple vision encoders for comprehensive video understanding.

Vision-language models (VLMs) are typically composed of a vision encoder, e.g. CLIP, and a language model (LM) that interprets the encoded features to solve downstream tasks. Despite remarkable progress, VLMs are subject to several shortcomings due to the limited capabilities of vision encoders, e.g. "blindness" to certain image features, visual hallucination, etc. To address these issues, we study broadening the visual encoding capabilities of VLMs. We first comprehensively benchmark several vision encoders with different inductive biases for solving VLM tasks. We observe that there is no single encoding configuration that consistently achieves top performance across different tasks, and encoders with different biases can perform surprisingly similarly. Motivated by this, we introduce a method, named BRAVE, that consolidates features from multiple frozen encoders into a more versatile representation that can be directly fed as the input to a frozen LM. BRAVE achieves state-of-the-art performance on a broad range of captioning and VQA benchmarks and significantly reduces the aforementioned issues of VLMs, while requiring a smaller number of trainable parameters than existing methods and having a more compressed representation. Our results highlight the potential of incorporating different visual biases for a more broad and contextualized visual understanding of VLMs.

Visual encoders are fundamental components in vision-language models (VLMs), each showcasing unique strengths derived from various pre-trained visual foundation models. To leverage the various capabilities of these encoders, recent studies incorporate multiple encoders within a single VLM, leading to a considerable increase in computational cost. In this paper, we present Mixture-of-Visual-Encoder Knowledge Distillation (MoVE-KD), a novel framework that distills the unique proficiencies of multiple vision encoders into a single, efficient encoder model. Specifically, to mitigate conflicts and retain the unique characteristics of each teacher encoder, we employ low-rank adaptation (LoRA) and mixture-of-experts (MoEs) to selectively activate specialized knowledge based on input features, enhancing both adaptability and efficiency. To regularize the KD process and enhance performance, we propose an attention-based distillation strategy that adaptively weighs the different visual encoders and emphasizes valuable visual tokens, reducing the burden of replicating comprehensive but distinct features from multiple teachers. Comprehensive experiments on popular VLMs, such as LLaVA and LLaVA-NeXT, validate the effectiveness of our method. The code will be released.

Large vision language models (LVLMs) integrate large language models (LLMs) with pre-trained vision encoders, thereby activating the perception capability of the model to understand image inputs for different queries and conduct subsequent reasoning. Improving this capability requires high-quality vision-language data, which is costly and labor-intensive to acquire. Self-training approaches have been effective in single-modal settings to alleviate the need for labeled data by leveraging model's own generation. However, effective self-training remains a challenge regarding the unique visual perception and reasoning capability of LVLMs. To address this, we introduce Self-Training on Image Comprehension (STIC), which emphasizes a self-training approach specifically for image comprehension. First, the model self-constructs a preference dataset for image descriptions using unlabeled images. Preferred responses are generated through a step-by-step prompt, while dis-preferred responses are generated from either corrupted images or misleading prompts. To further self-improve reasoning on the extracted visual information, we let the model reuse a small portion of existing instruction-tuning data and append its self-generated image descriptions to the prompts. We validate the effectiveness of STIC across seven different benchmarks, demonstrating substantial performance gains of 4.0% on average while using 70% less supervised fine-tuning data than the current method. Further studies investigate various components of STIC and highlight its potential to leverage vast quantities of unlabeled images for self-training. Code and data are made publicly available.

Existing encoder-free vision-language models (VLMs) are rapidly narrowing the performance gap with their encoder-based counterparts, highlighting the promising potential for unified multimodal systems with structural simplicity and efficient deployment. We systematically clarify the performance gap between VLMs using pre-trained vision encoders, discrete tokenizers, and minimalist visual layers from scratch, deeply excavating the under-examined characteristics of encoder-free VLMs. We develop efficient strategies for encoder-free VLMs that rival mainstream encoder-based ones. After an in-depth investigation, we launch EVEv2.0, a new and improved family of encoder-free VLMs. We show that: (i) Properly decomposing and hierarchically associating vision and language within a unified model reduces interference between modalities. (ii) A well-designed training strategy enables effective optimization for encoder-free VLMs. Through extensive evaluation, our EVEv2.0 represents a thorough study for developing a decoder-only architecture across modalities, demonstrating superior data efficiency and strong vision-reasoning capability. Code is publicly available at: https://github.com/baaivision/EVE.

Vision language models (VLMs) have drastically changed the computer vision model landscape in only a few years, opening an exciting array of new applications from zero-shot image classification, over to image captioning, and visual question answering. Unlike pure vision models, they offer an intuitive way to access visual content through language prompting. The wide applicability of such models encourages us to ask whether they also align with human vision - specifically, how far they adopt human-induced visual biases through multimodal fusion, or whether they simply inherit biases from pure vision models. One important visual bias is the texture vs. shape bias, or the dominance of local over global information. In this paper, we study this bias in a wide range of popular VLMs. Interestingly, we find that VLMs are often more shape-biased than their vision encoders, indicating that visual biases are modulated to some extent through text in multimodal models. If text does indeed influence visual biases, this suggests that we may be able to steer visual biases not just through visual input but also through language: a hypothesis that we confirm through extensive experiments. For instance, we are able to steer shape bias from as low as 49% to as high as 72% through prompting alone. For now, the strong human bias towards shape (96%) remains out of reach for all tested VLMs.

Recent Multimodal Large Language Models (MLLMs) exhibit impressive abilities to perceive images and follow open-ended instructions. The capabilities of MLLMs depend on two crucial factors: the model architecture to facilitate the feature alignment of visual modules and large language models; the multimodal instruction tuning datasets for human instruction following. (i) For the model architecture, most existing models introduce an external bridge module to connect vision encoders with language models, which needs an additional feature-alignment pre-training. In this work, we discover that compact pre-trained vision language models can inherently serve as ``out-of-the-box'' bridges between vision and language. Based on this, we propose Muffin framework, which directly employs pre-trained vision-language models to act as providers of visual signals. (ii) For the multimodal instruction tuning datasets, existing methods omit the complementary relationship between different datasets and simply mix datasets from different tasks. Instead, we propose UniMM-Chat dataset which explores the complementarities of datasets to generate 1.1M high-quality and diverse multimodal instructions. We merge information describing the same image from diverse datasets and transforms it into more knowledge-intensive conversation data. Experimental results demonstrate the effectiveness of the Muffin framework and UniMM-Chat dataset. Muffin achieves state-of-the-art performance on a wide range of vision-language tasks, significantly surpassing state-of-the-art models like LLaVA and InstructBLIP. Our model and dataset are all accessible at https://github.com/thunlp/muffin.

We present Florence-VL, a new family of multimodal large language models (MLLMs) with enriched visual representations produced by Florence-2, a generative vision foundation model. Unlike the widely used CLIP-style vision transformer trained by contrastive learning, Florence-2 can capture different levels and aspects of visual features, which are more versatile to be adapted to diverse downstream tasks. We propose a novel feature-fusion architecture and an innovative training recipe that effectively integrates Florence-2's visual features into pretrained LLMs, such as Phi 3.5 and LLama 3. In particular, we propose "depth-breath fusion (DBFusion)" to fuse the visual features extracted from different depths and under multiple prompts. Our model training is composed of end-to-end pretraining of the whole model followed by finetuning of the projection layer and the LLM, on a carefully designed recipe of diverse open-source datasets that include high-quality image captions and instruction-tuning pairs. Our quantitative analysis and visualization of Florence-VL's visual features show its advantages over popular vision encoders on vision-language alignment, where the enriched depth and breath play important roles. Florence-VL achieves significant improvements over existing state-of-the-art MLLMs across various multi-modal and vision-centric benchmarks covering general VQA, perception, hallucination, OCR, Chart, knowledge-intensive understanding, etc. To facilitate future research, our models and the complete training recipe are open-sourced. https://github.com/JiuhaiChen/Florence-VL

This paper demonstrates that a progressively aligned language model can effectively bridge frozen vision encoders and large language models (LLMs). While the fundamental architecture and pre-training methods of vision encoders and LLMs have been extensively studied, the architecture and training strategy of vision-language adapters vary significantly across recent works. Our research undertakes a thorough exploration of the state-of-the-art perceiver resampler architecture and builds a strong baseline. However, we observe that the vision-language alignment with perceiver resampler exhibits slow convergence and limited scalability with a lack of direct supervision. To address this issue, we propose PaLM2-VAdapter, employing a progressively aligned language model as the vision-language adapter. Compared to the strong baseline with perceiver resampler, our method empirically shows faster convergence, higher performance, and stronger scalability. Extensive experiments across various Visual Question Answering (VQA) and captioning tasks on both images and videos demonstrate that our model exhibits state-of-the-art visual understanding and multi-modal reasoning capabilities. Notably, our method achieves these advancements with 30~70% fewer parameters than the state-of-the-art large vision-language models, marking a significant efficiency improvement.

Recently, Vision Large Language Models (VLLMs) integrated with vision encoders have shown promising performance in vision understanding. The key of VLLMs is to encode visual content into sequences of visual tokens, enabling VLLMs to simultaneously process both visual and textual content. However, understanding videos, especially long videos, remain a challenge to VLLMs as the number of visual tokens grows rapidly when encoding videos, resulting in the risk of exceeding the context window of VLLMs and introducing heavy computation burden. To restrict the number of visual tokens, existing VLLMs either: (1) uniformly downsample videos into a fixed number of frames or (2) reducing the number of visual tokens encoded from each frame. We argue the former solution neglects the rich temporal cue in videos and the later overlooks the spatial details in each frame. In this work, we present Balanced-VLLM (B-VLLM): a novel VLLM framework that aims to effectively leverage task relevant spatio-temporal cues while restricting the number of visual tokens under the VLLM context window length. At the core of our method, we devise a text-conditioned adaptive frame selection module to identify frames relevant to the visual understanding task. The selected frames are then de-duplicated using a temporal frame token merging technique. The visual tokens of the selected frames are processed through a spatial token sampling module and an optional spatial token merging strategy to achieve precise control over the token count. Experimental results show that B-VLLM is effective in balancing the number of frames and visual tokens in video understanding, yielding superior performance on various video understanding benchmarks. Our code is available at https://github.com/zhuqiangLu/B-VLLM.

How well are unimodal vision and language models aligned? Although prior work have approached answering this question, their assessment methods do not directly translate to how these models are used in practical vision-language tasks. In this paper, we propose a direct assessment method, inspired by linear probing, to assess vision-language alignment. We identify that the degree of alignment of the SSL vision models depends on their SSL training objective, and we find that the clustering quality of SSL representations has a stronger impact on alignment performance than their linear separability. Next, we introduce Swift Alignment of Image and Language (SAIL), a efficient transfer learning framework that aligns pretrained unimodal vision and language models for downstream vision-language tasks. Since SAIL leverages the strengths of pretrained unimodal models, it requires significantly fewer (6%) paired image-text data for the multimodal alignment compared to models like CLIP which are trained from scratch. SAIL training only requires a single A100 GPU, 5 hours of training and can accommodate a batch size up to 32,768. SAIL achieves 73.4% zero-shot accuracy on ImageNet (vs. CLIP's 72.7%) and excels in zero-shot retrieval, complex reasoning, and semantic segmentation. Additionally, SAIL improves the language-compatibility of vision encoders that in turn enhance the performance of multimodal large language models. The entire codebase and model weights are open-source: https://lezhang7.github.io/sail.github.io/

Despite their promise to perform complex reasoning, large language models (LLMs) have been shown to have limited effectiveness in end-to-end planning. This has inspired an intriguing question: if these models cannot plan well, can they still contribute to the planning framework as a helpful plan evaluator? In this work, we generalize this question to consider LLMs augmented with visual understanding, i.e., Vision-Language Models (VLMs). We introduce PathEval, a novel benchmark evaluating VLMs as plan evaluators in complex path-planning scenarios. Succeeding in the benchmark requires a VLM to be able to abstract traits of optimal paths from the scenario description, demonstrate precise low-level perception on each path, and integrate this information to decide the better path. Our analysis of state-of-the-art VLMs reveals that these models face significant challenges on the benchmark. We observe that the VLMs can precisely abstract given scenarios to identify the desired traits and exhibit mixed performance in integrating the provided information. Yet, their vision component presents a critical bottleneck, with models struggling to perceive low-level details about a path. Our experimental results show that this issue cannot be trivially addressed via end-to-end fine-tuning; rather, task-specific discriminative adaptation of these vision encoders is needed for these VLMs to become effective path evaluators.

This work presents ParGo, a novel Partial-Global projector designed to connect the vision and language modalities for Multimodal Large Language Models (MLLMs). Unlike previous works that rely on global attention-based projectors, our ParGo bridges the representation gap between the separately pre-trained vision encoders and the LLMs by integrating global and partial views, which alleviates the overemphasis on prominent regions. To facilitate the effective training of ParGo, we collect a large-scale detail-captioned image-text dataset named ParGoCap-1M-PT, consisting of 1 million images paired with high-quality captions. Extensive experiments on several MLLM benchmarks demonstrate the effectiveness of our ParGo, highlighting its superiority in aligning vision and language modalities. Compared to conventional Q-Former projector, our ParGo achieves an improvement of 259.96 in MME benchmark. Furthermore, our experiments reveal that ParGo significantly outperforms other projectors, particularly in tasks that emphasize detail perception ability.

Medical vision-and-language pre-training (Med-VLP) has received considerable attention owing to its applicability to extracting generic vision-and-language representations from medical images and texts. Most existing methods mainly contain three elements: uni-modal encoders (i.e., a vision encoder and a language encoder), a multi-modal fusion module, and pretext tasks, with few studies considering the importance of medical domain expert knowledge and explicitly exploiting such knowledge to facilitate Med-VLP. Although there exist knowledge-enhanced vision-and-language pre-training (VLP) methods in the general domain, most require off-the-shelf toolkits (e.g., object detectors and scene graph parsers), which are unavailable in the medical domain. In this paper, we propose a systematic and effective approach to enhance Med-VLP by structured medical knowledge from three perspectives. First, considering knowledge can be regarded as the intermediate medium between vision and language, we align the representations of the vision encoder and the language encoder through knowledge. Second, we inject knowledge into the multi-modal fusion model to enable the model to perform reasoning using knowledge as the supplementation of the input image and text. Third, we guide the model to put emphasis on the most critical information in images and texts by designing knowledge-induced pretext tasks. To perform a comprehensive evaluation and facilitate further research, we construct a medical vision-and-language benchmark including three tasks. Experimental results illustrate the effectiveness of our approach, where state-of-the-art performance is achieved on all downstream tasks. Further analyses explore the effects of different components of our approach and various settings of pre-training.

Contrastive pretraining on image-text pairs from the web is one of the most popular large-scale pretraining strategies for vision backbones, especially in the context of large multimodal models. At the same time, image captioning on this type of data is commonly considered an inferior pretraining strategy. In this paper, we perform a fair comparison of these two pretraining strategies, carefully matching training data, compute, and model capacity. Using a standard encoder-decoder transformer, we find that captioning alone is surprisingly effective: on classification tasks, captioning produces vision encoders competitive with contrastively pretrained encoders, while surpassing them on vision & language tasks. We further analyze the effect of the model architecture and scale, as well as the pretraining data on the representation quality, and find that captioning exhibits the same or better scaling behavior along these axes. Overall our results show that plain image captioning is a more powerful pretraining strategy than was previously believed.

Large multimodal models (LMMs) have achieved impressive progress in vision-language understanding, yet they face limitations in real-world applications requiring complex reasoning over a large number of images. Existing benchmarks for multi-image question-answering are limited in scope, each question is paired with only up to 30 images, which does not fully capture the demands of large-scale retrieval tasks encountered in the real-world usages. To reduce these gaps, we introduce two document haystack benchmarks, dubbed DocHaystack and InfoHaystack, designed to evaluate LMM performance on large-scale visual document retrieval and understanding. Additionally, we propose V-RAG, a novel, vision-centric retrieval-augmented generation (RAG) framework that leverages a suite of multimodal vision encoders, each optimized for specific strengths, and a dedicated question-document relevance module. V-RAG sets a new standard, with a 9% and 11% improvement in Recall@1 on the challenging DocHaystack-1000 and InfoHaystack-1000 benchmarks, respectively, compared to the previous best baseline models. Additionally, integrating V-RAG with LMMs enables them to efficiently operate across thousands of images, yielding significant improvements on our DocHaystack and InfoHaystack benchmarks. Our code and datasets are available at https://github.com/Vision-CAIR/dochaystacks

An increasing number of vision-language tasks can be handled with little to no training, i.e., in a zero and few-shot manner, by marrying large language models (LLMs) to vision encoders, resulting in large vision-language models (LVLMs). While this has huge upsides, such as not requiring training data or custom architectures, how an input is presented to a LVLM can have a major impact on zero-shot model performance. In particular, inputs phrased in an underspecified way can result in incorrect answers due to factors like missing visual information, complex implicit reasoning, or linguistic ambiguity. Therefore, adding visually grounded information to the input as a preemptive clarification should improve model performance by reducing underspecification, e.g., by localizing objects and disambiguating references. Similarly, in the VQA setting, changing the way questions are framed can make them easier for models to answer. To this end, we present Rephrase, Augment and Reason (RepARe), a gradient-free framework that extracts salient details about the image using the underlying LVLM as a captioner and reasoner, in order to propose modifications to the original question. We then use the LVLM's confidence over a generated answer as an unsupervised scoring function to select the rephrased question most likely to improve zero-shot performance. Focusing on two visual question answering tasks, we show that RepARe can result in a 3.85% (absolute) increase in zero-shot performance on VQAv2 and a 6.41% point increase on A-OKVQA. Additionally, we find that using gold answers for oracle question candidate selection achieves a substantial gain in VQA accuracy by up to 14.41%. Through extensive analysis, we demonstrate that outputs from RepARe increase syntactic complexity, and effectively utilize vision-language interaction and the frozen language model in LVLMs.

The recently proposed SparseFormer architecture provides an alternative approach to visual understanding by utilizing a significantly lower number of visual tokens via adjusting RoIs, greatly reducing computational costs while still achieving promising performance. However, training SparseFormers from scratch is still expensive, and scaling up the number of parameters can be challenging. In this paper, we propose to bootstrap SparseFormers from ViT-based vision foundation models in a simple and efficient way. Since the majority of SparseFormer blocks are the standard transformer ones, we can inherit weights from large-scale pre-trained vision transformers and freeze them as much as possible. Therefore, we only need to train the SparseFormer-specific lightweight focusing transformer to adjust token RoIs and fine-tune a few early pre-trained blocks to align the final token representation. In such a way, we can bootstrap SparseFormer architectures from various large-scale pre-trained models (e.g., IN-21K pre-trained AugRegs or CLIPs) using a rather smaller amount of training samples (e.g., IN-1K) and without labels or captions within just a few hours. As a result, the bootstrapped unimodal SparseFormer (from AugReg-ViT-L/16-384) can reach 84.9% accuracy on IN-1K with only 49 tokens, and the multimodal SparseFormer from CLIPs also demonstrates notable zero-shot performance with highly reduced computational cost without seeing any caption during the bootstrapping procedure. In addition, CLIP-bootstrapped SparseFormers, which align the output space with language without seeing a word, can serve as efficient vision encoders in multimodal large language models. Code will be publicly available at https://github.com/showlab/sparseformer

Recent advancements in multi-modal models have significantly improved vision-language alignment in radiology. However, existing approaches struggle to effectively utilize complex radiology reports for learning, rely on low-resolution images, and offer limited interpretability in attention mechanisms. To address these challenges, we introduce RadZero, a novel similarity-based cross-attention framework for vision-language alignment in radiology with zero-shot multi-task capability. RadZero leverages large language models to extract minimal semantic sentences from radiology reports and employs a multi-positive contrastive learning strategy to effectively capture relationships between images and multiple relevant textual descriptions. It also utilizes a pre-trained vision encoder with additional trainable Transformer layers, allowing efficient high-resolution image processing. By computing similarity between text embeddings and local image patch features, RadZero enables zero-shot inference with similarity probability for classification and pixel-level cross-modal similarity maps for grounding and segmentation. Experimental results on public chest radiograph benchmarks show that RadZero outperforms state-of-the-art methods in zero-shot classification, grounding, and segmentation. Furthermore, cross-modal similarity map analysis highlights its potential for improving explainability in vision-language alignment. Additionally, qualitative evaluation demonstrates RadZero's capability for open-vocabulary semantic segmentation, further validating its effectiveness in medical imaging.

Recent advancements in vision-language models have enhanced performance by increasing the length of visual tokens, making them much longer than text tokens and significantly raising computational costs. However, we observe that the visual tokens generated by popular vision encoders, such as CLIP and SigLIP, contain significant redundancy. To address this, we introduce VisionZip, a simple yet effective method that selects a set of informative tokens for input to the language model, reducing visual token redundancy and improving efficiency while maintaining model performance. The proposed VisionZip can be widely applied to image and video understanding tasks and is well-suited for multi-turn dialogues in real-world scenarios, where previous methods tend to underperform. Experimental results show that VisionZip outperforms the previous state-of-the-art method by at least 5% performance gains across nearly all settings. Moreover, our method significantly enhances model inference speed, improving the prefilling time by 8x and enabling the LLaVA-Next 13B model to infer faster than the LLaVA-Next 7B model while achieving better results. Furthermore, we analyze the causes of this redundancy and encourage the community to focus on extracting better visual features rather than merely increasing token length. Our code is available at https://github.com/dvlab-research/VisionZip .

The rapid development of large language and vision models (LLVMs) has been driven by advances in visual instruction tuning. Recently, open-source LLVMs have curated high-quality visual instruction tuning datasets and utilized additional vision encoders or multiple computer vision models in order to narrow the performance gap with powerful closed-source LLVMs. These advancements are attributed to multifaceted information required for diverse capabilities, including fundamental image understanding, real-world knowledge about common-sense and non-object concepts (e.g., charts, diagrams, symbols, signs, and math problems), and step-by-step procedures for solving complex questions. Drawing from the multifaceted information, we present a new efficient LLVM, Mamba-based traversal of rationales (Meteor), which leverages multifaceted rationale to enhance understanding and answering capabilities. To embed lengthy rationales containing abundant information, we employ the Mamba architecture, capable of processing sequential data with linear time complexity. We introduce a new concept of traversal of rationale that facilitates efficient embedding of rationale. Subsequently, the backbone multimodal language model (MLM) is trained to generate answers with the aid of rationale. Through these steps, Meteor achieves significant improvements in vision language performances across multiple evaluation benchmarks requiring diverse capabilities, without scaling up the model size or employing additional vision encoders and computer vision models.

We introduce Cambrian-1, a family of multimodal LLMs (MLLMs) designed with a vision-centric approach. While stronger language models can enhance multimodal capabilities, the design choices for vision components are often insufficiently explored and disconnected from visual representation learning research. This gap hinders accurate sensory grounding in real-world scenarios. Our study uses LLMs and visual instruction tuning as an interface to evaluate various visual representations, offering new insights into different models and architectures -- self-supervised, strongly supervised, or combinations thereof -- based on experiments with over 20 vision encoders. We critically examine existing MLLM benchmarks, addressing the difficulties involved in consolidating and interpreting results from various tasks, and introduce a new vision-centric benchmark, CV-Bench. To further improve visual grounding, we propose the Spatial Vision Aggregator (SVA), a dynamic and spatially-aware connector that integrates high-resolution vision features with LLMs while reducing the number of tokens. Additionally, we discuss the curation of high-quality visual instruction-tuning data from publicly available sources, emphasizing the importance of data source balancing and distribution ratio. Collectively, Cambrian-1 not only achieves state-of-the-art performance but also serves as a comprehensive, open cookbook for instruction-tuned MLLMs. We provide model weights, code, supporting tools, datasets, and detailed instruction-tuning and evaluation recipes. We hope our release will inspire and accelerate advancements in multimodal systems and visual representation learning.

This work focuses on the potential of Vision LLMs (VLLMs) in visual reasoning. Different from prior studies, we shift our focus from evaluating standard performance to introducing a comprehensive safety evaluation suite, covering both out-of-distribution (OOD) generalization and adversarial robustness. For the OOD evaluation, we present two novel VQA datasets, each with one variant, designed to test model performance under challenging conditions. In exploring adversarial robustness, we propose a straightforward attack strategy for misleading VLLMs to produce visual-unrelated responses. Moreover, we assess the efficacy of two jailbreaking strategies, targeting either the vision or language component of VLLMs. Our evaluation of 21 diverse models, ranging from open-source VLLMs to GPT-4V, yields interesting observations: 1) Current VLLMs struggle with OOD texts but not images, unless the visual information is limited; and 2) These VLLMs can be easily misled by deceiving vision encoders only, and their vision-language training often compromise safety protocols. We release this safety evaluation suite at https://github.com/UCSC-VLAA/vllm-safety-benchmark.

Vision encoders typically generate a large number of visual tokens, providing information-rich representations but significantly increasing computational demands. This raises the question of whether all generated tokens are equally valuable or if some of them can be discarded to reduce computational costs without compromising quality. In this paper, we introduce a new method for determining feature utility based on the idea that less valuable features can be reconstructed from more valuable ones. We implement this concept by integrating an autoencoder with a Gumbel-Softmax selection mechanism, that allows identifying and retaining only the most informative visual tokens. To validate our approach, we compared the performance of the LLaVA-NeXT model, using features selected by our method with randomly selected features. We found that on OCR-based tasks, more than 50% of the visual context can be removed with minimal performance loss, whereas randomly discarding the same proportion of features significantly affects the model capabilities. Furthermore, in general-domain tasks, even randomly retaining only 30% of tokens achieves performance comparable to using the full set of visual tokens. Our results highlight a promising direction towards adaptive and efficient multimodal pruning that facilitates scalable and low-overhead inference without compromising performance.

Vision encoders are increasingly used in modern applications, from vision-only models to multimodal systems such as vision-language models. Despite their remarkable success, it remains unclear how these architectures represent features internally. Here, we propose a novel approach for interpreting vision features via image reconstruction. We compare two related model families, SigLIP and SigLIP2, which differ only in their training objective, and show that encoders pre-trained on image-based tasks retain significantly more image information than those trained on non-image tasks such as contrastive learning. We further apply our method to a range of vision encoders, ranking them by the informativeness of their feature representations. Finally, we demonstrate that manipulating the feature space yields predictable changes in reconstructed images, revealing that orthogonal rotations (rather than spatial transformations) control color encoding. Our approach can be applied to any vision encoder, shedding light on the inner structure of its feature space. The code and model weights to reproduce the experiments are available in GitHub.

Do we fully leverage the potential of visual encoder in Multimodal Large Language Models (MLLMs)? The recent outstanding performance of MLLMs in multimodal understanding has garnered broad attention from both academia and industry. In the current MLLM rat race, the focus seems to be predominantly on the linguistic side. We witness the rise of larger and higher-quality instruction datasets, as well as the involvement of larger-sized LLMs. Yet, scant attention has been directed towards the visual signals utilized by MLLMs, often assumed to be the final high-level features extracted by a frozen visual encoder. In this paper, we introduce the Dense Connector - a simple, effective, and plug-and-play vision-language connector that significantly enhances existing MLLMs by leveraging multi-layer visual features, with minimal additional computational overhead. Furthermore, our model, trained solely on images, showcases remarkable zero-shot capabilities in video understanding as well. Experimental results across various vision encoders, image resolutions, training dataset scales, varying sizes of LLMs (2.7B->70B), and diverse architectures of MLLMs (e.g., LLaVA and Mini-Gemini) validate the versatility and scalability of our approach, achieving state-of-the-art performance on across 19 image and video benchmarks. We hope that this work will provide valuable experience and serve as a basic module for future MLLM development.

Generating semantically and temporally aligned audio content in accordance with video input has become a focal point for researchers, particularly following the remarkable breakthrough in text-to-video generation. In this work, we aim to offer insights into the video-to-audio generation paradigm, focusing on three crucial aspects: vision encoders, auxiliary embeddings, and data augmentation techniques. Beginning with a foundational model VTA-LDM built on a simple yet surprisingly effective intuition, we explore various vision encoders and auxiliary embeddings through ablation studies. Employing a comprehensive evaluation pipeline that emphasizes generation quality and video-audio synchronization alignment, we demonstrate that our model exhibits state-of-the-art video-to-audio generation capabilities. Furthermore, we provide critical insights into the impact of different data augmentation methods on enhancing the generation framework's overall capacity. We showcase possibilities to advance the challenge of generating synchronized audio from semantic and temporal perspectives. We hope these insights will serve as a stepping stone toward developing more realistic and accurate audio-visual generation models.

Most existing vision encoders map images into a fixed-length sequence of tokens, overlooking the fact that different images contain varying amounts of information. For example, a visually complex image (e.g., a cluttered room) inherently carries more information and thus deserves more tokens than a simple image (e.g., a blank wall). To address this inefficiency, we propose DOVE, a dynamic vision encoder that produces a variable number of visual tokens (i.e., continuous representation vectors) to reconstruct each image. Our results show that DOVE significantly reduces the average number of tokens while maintaining high reconstruction quality. In several linear probing and downstream multimodal tasks, it outperforms existing autoencoder-based tokenization methods when using far fewer tokens, capturing more expressive semantic features compared to fixed-length encoding. We further extend DOVE with query-conditioned tokenization. By guiding the model to focus on query-relevant regions, it achieves more efficient and targeted semantic extraction. Our code and checkpoints are available at https://dove-encoder.github.io/dove-encoder.

The field of visual representation learning has seen explosive growth in the past years, but its benefits in robotics have been surprisingly limited so far. Prior work uses generic visual representations as a basis to learn (task-specific) robot action policies (e.g., via behavior cloning). While the visual representations do accelerate learning, they are primarily used to encode visual observations. Thus, action information has to be derived purely from robot data, which is expensive to collect! In this work, we present a scalable alternative where the visual representations can help directly infer robot actions. We observe that vision encoders express relationships between image observations as distances (e.g., via embedding dot product) that could be used to efficiently plan robot behavior. We operationalize this insight and develop a simple algorithm for acquiring a distance function and dynamics predictor, by fine-tuning a pre-trained representation on human collected video sequences. The final method is able to substantially outperform traditional robot learning baselines (e.g., 70% success v.s. 50% for behavior cloning on pick-place) on a suite of diverse real-world manipulation tasks. It can also generalize to novel objects, without using any robot demonstrations during train time. For visualizations of the learned policies please check: https://agi-labs.github.io/manipulate-by-seeing/.

The rapid development of Multimodal Large Language Models (MLLMs) like GPT-4V has marked a significant step towards artificial general intelligence. Existing methods mainly focus on aligning vision encoders with LLMs through supervised fine-tuning (SFT) to endow LLMs with multimodal abilities, making MLLMs' inherent ability to react to multiple languages progressively deteriorate as the training process evolves. We empirically find that the imbalanced SFT datasets, primarily composed of English-centric image-text pairs, lead to significantly reduced performance in non-English languages. This is due to the failure of aligning the vision encoder and LLM with multilingual tokens during the SFT process. In this paper, we introduce Parrot, a novel method that utilizes textual guidance to drive visual token alignment at the language level. Parrot makes the visual tokens condition on diverse language inputs and uses Mixture-of-Experts (MoE) to promote the alignment of multilingual tokens. Specifically, to enhance non-English visual tokens alignment, we compute the cross-attention using the initial visual features and textual embeddings, the result of which is then fed into the MoE router to select the most relevant experts. The selected experts subsequently convert the initial visual tokens into language-specific visual tokens. Moreover, considering the current lack of benchmarks for evaluating multilingual capabilities within the field, we collect and make available a Massive Multilingual Multimodal Benchmark which includes 6 languages, 15 categories, and 12,000 questions, named as MMMB. Our method not only demonstrates state-of-the-art performance on multilingual MMBench and MMMB, but also excels across a broad range of multimodal tasks. Both the source code and the training dataset of Parrot will be made publicly available.

We present the TinyLLaVA framework that provides a unified perspective in designing and analyzing the small-scale Large Multimodal Models (LMMs). We empirically study the effects of different vision encoders, connection modules, language models, training data and training recipes. Our extensive experiments showed that better quality of data combined with better training recipes, smaller LMMs can consistently achieve on-par performances compared to bigger LMMs. Under our framework, we train a family of small-scale LMMs. Our best model, TinyLLaVA-3.1B, achieves better overall performance against existing 7B models such as LLaVA-1.5 and Qwen-VL. We hope our findings can serve as baselines for future research in terms of data scaling, training setups and model selections. Our model weights and codes will be made public.

Image representations are often evaluated through disjointed, task-specific protocols, leading to a fragmented understanding of model capabilities. For instance, it is unclear whether an image embedding model adept at clustering images is equally good at retrieving relevant images given a piece of text. We introduce the Massive Image Embedding Benchmark (MIEB) to evaluate the performance of image and image-text embedding models across the broadest spectrum to date. MIEB spans 38 languages across 130 individual tasks, which we group into 8 high-level categories. We benchmark 50 models across our benchmark, finding that no single method dominates across all task categories. We reveal hidden capabilities in advanced vision models such as their accurate visual representation of texts, and their yet limited capabilities in interleaved encodings and matching images and texts in the presence of confounders. We also show that the performance of vision encoders on MIEB correlates highly with their performance when used in multimodal large language models. Our code, dataset, and leaderboard are publicly available at https://github.com/embeddings-benchmark/mteb.

The exponential growth of Large Multimodal Models (LMMs) has driven advancements in cross-modal reasoning but at significant computational costs. In this work, we focus on visual language models. We highlight the redundancy and inefficiency in current vision encoders, and seek to construct an adaptive compression method for multimodal data. In this work, we characterize a panoply of visual token selection and merging approaches through both benchmarking and qualitative analysis. In particular, we demonstrate that simple cluster-level token aggregation outperforms prior state-of-the-art works in token selection and merging, including merging at the vision encoder level and attention-based approaches. We underline the redundancy in current vision encoders, and shed light on several puzzling trends regarding principles of visual token selection through cross-modal attention visualizations. This work is a first effort towards more effective encoding and processing of high-dimensional data, and paves the way for more scalable and sustainable multimodal systems.

We introduce Perception Encoder (PE), a state-of-the-art encoder for image and video understanding trained via simple vision-language learning. Traditionally, vision encoders have relied on a variety of pretraining objectives, each tailored to specific downstream tasks such as classification, captioning, or localization. Surprisingly, after scaling our carefully tuned image pretraining recipe and refining with our robust video data engine, we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the intermediate layers of the network. To draw them out, we introduce two alignment methods, language alignment for multimodal language modeling, and spatial alignment for dense prediction. Together with the core contrastive checkpoint, our PE family of models achieves state-of-the-art performance on a wide variety of tasks, including zero-shot image and video classification and retrieval; document, image, and video Q&A; and spatial tasks such as detection, depth estimation, and tracking. To foster further research, we are releasing our models, code, and a novel dataset of synthetically and human-annotated videos.

In this paper, we present OtterHD-8B, an innovative multimodal model evolved from Fuyu-8B, specifically engineered to interpret high-resolution visual inputs with granular precision. Unlike conventional models that are constrained by fixed-size vision encoders, OtterHD-8B boasts the ability to handle flexible input dimensions, ensuring its versatility across various inference requirements. Alongside this model, we introduce MagnifierBench, an evaluation framework designed to scrutinize models' ability to discern minute details and spatial relationships of small objects. Our comparative analysis reveals that while current leading models falter on this benchmark, OtterHD-8B, particularly when directly processing high-resolution inputs, outperforms its counterparts by a substantial margin. The findings illuminate the structural variances in visual information processing among different models and the influence that the vision encoders' pre-training resolution disparities have on model effectiveness within such benchmarks. Our study highlights the critical role of flexibility and high-resolution input capabilities in large multimodal models and also exemplifies the potential inherent in the Fuyu architecture's simplicity for handling complex visual data.

The dominance of large decoder-only language models has overshadowed encoder-decoder architectures, despite their fundamental efficiency advantages in sequence processing. For small language models (SLMs) - those with 1 billion parameters or fewer - our systematic analysis across GPU, CPU, and NPU platforms reveals that encoder-decoder architectures achieve 47% lower first-token latency and 4.7x higher throughput compared to decoder-only models on edge devices. These gains may be attributed to encoder-decoder's one-time input processing and efficient separation of understanding and generation phases. We introduce a novel knowledge distillation framework that enables encoder-decoder models to leverage capabilities from large scalable decoder-only teachers while preserving their architectural advantages, achieving up to 6 average performance points improvement across diverse tasks, with significant gains in asymmetric sequence tasks where input and output distributions can benefit from different processing approaches. When combined with modern advances like Rotary Positional Embeddings (RoPE) and Vision encoders, our systematic investigation demonstrates that encoder-decoder architectures provide a more practical path toward deploying capable language models in resource-constrained environments. Our findings challenge the prevailing trend toward decoder-only scaling, showing that architectural choices become increasingly crucial as parameter budgets decrease, particularly for on-device and edge deployments where computational efficiency is paramount.

Knowledge distillation (KD) is the de facto standard for compressing large-scale models into smaller ones. Prior works have explored ever more complex KD strategies involving different objective functions, teacher-ensembles, and weight inheritance. In this work we explore an alternative, yet simple approach -- active data curation as effective distillation for contrastive multimodal pretraining. Our simple online batch selection method, ACID, outperforms strong KD baselines across various model-, data- and compute-configurations. Further, we find such an active data curation strategy to in fact be complementary to standard KD, and can be effectively combined to train highly performant inference-efficient models. Our simple and scalable pretraining framework, ACED, achieves state-of-the-art results across 27 zero-shot classification and retrieval tasks with upto 11% less inference FLOPs. We further demonstrate that our ACED models yield strong vision-encoders for training generative multimodal models in the LiT-Decoder setting, outperforming larger vision encoders for image-captioning and visual question-answering tasks.

Previous works show that noisy, web-crawled image-text pairs may limit vision-language pretraining like CLIP and propose learning with synthetic captions as a promising alternative. Our work continues this effort, introducing two simple yet effective designs to better leverage richly described synthetic captions. Firstly, by observing a strong inverse effect in learning with synthetic captions -- the short synthetic captions can generally lead to MUCH higher performance than full-length ones -- we therefore fed only partial synthetic captions to the text encoder. Secondly, we incorporate an autoregressive captioner to mimic the recaptioning process -- by conditioning on the paired image input and web-crawled text description, the captioner learns to predict the full-length synthetic caption generated by advanced MLLMs. Experiments show that our framework significantly improves zero-shot performance in cross-modal retrieval tasks, setting new SOTA results on MSCOCO and Flickr30K. Moreover, such trained vision encoders can enhance the visual capability of LLaVA, showing strong improvements on a range of MLLM benchmarks. Our project page is https://ucsc-vlaa.github.io/CLIPS/.

Foundation models have received much attention due to their effectiveness across a broad range of downstream applications. Though there is a big convergence in terms of architecture, most pretrained models are typically still developed for specific tasks or modalities. In this work, we propose to use language models as a general-purpose interface to various foundation models. A collection of pretrained encoders perceive diverse modalities (such as vision, and language), and they dock with a language model that plays the role of a universal task layer. We propose a semi-causal language modeling objective to jointly pretrain the interface and the modular encoders. We subsume the advantages and capabilities from both causal and non-causal modeling, thereby combining the best of two worlds. Specifically, the proposed method not only inherits the capabilities of in-context learning and open-ended generation from causal language modeling, but also is conducive to finetuning because of the bidirectional encoders. More importantly, our approach seamlessly unlocks the combinations of the above capabilities, e.g., enabling in-context learning or instruction following with finetuned encoders. Experimental results across various language-only and vision-language benchmarks show that our model outperforms or is competitive with specialized models on finetuning, zero-shot generalization, and few-shot learning.

The development of Multi-modal Large Language Models (MLLMs) enhances Large Language Models (LLMs) with the ability to perceive data formats beyond text, significantly advancing a range of downstream applications, such as visual question answering and image captioning. However, the substantial computational costs associated with processing high-resolution images and videos pose a barrier to their broader adoption. To address this challenge, compressing vision tokens in MLLMs has emerged as a promising approach to reduce inference costs. While existing methods conduct token reduction in the feature alignment phase. In this paper, we introduce VisToG, a novel grouping mechanism that leverages the capabilities of pre-trained vision encoders to group similar image segments without the need for segmentation masks. Specifically, we concatenate semantic tokens to represent image semantic segments after the linear projection layer before feeding into the vision encoder. Besides, with the isolated attention we adopt, VisToG can identify and eliminate redundant visual tokens utilizing the prior knowledge in the pre-trained vision encoder, which effectively reduces computational demands. Extensive experiments demonstrate the effectiveness of VisToG, maintaining 98.1% of the original performance while achieving a reduction of over 27\% inference time.

Color plays an important role in human perception and usually provides critical clues in visual reasoning. However, it is unclear whether and how vision-language models (VLMs) can perceive, understand, and leverage color as humans. This paper introduces ColorBench, an innovative benchmark meticulously crafted to assess the capabilities of VLMs in color understanding, including color perception, reasoning, and robustness. By curating a suite of diverse test scenarios, with grounding in real applications, ColorBench evaluates how these models perceive colors, infer meanings from color-based cues, and maintain consistent performance under varying color transformations. Through an extensive evaluation of 32 VLMs with varying language models and vision encoders, our paper reveals some undiscovered findings: (i) The scaling law (larger models are better) still holds on ColorBench, while the language model plays a more important role than the vision encoder. (ii) However, the performance gaps across models are relatively small, indicating that color understanding has been largely neglected by existing VLMs. (iii) CoT reasoning improves color understanding accuracies and robustness, though they are vision-centric tasks. (iv) Color clues are indeed leveraged by VLMs on ColorBench but they can also mislead models in some tasks. These findings highlight the critical limitations of current VLMs and underscore the need to enhance color comprehension. Our ColorBenchcan serve as a foundational tool for advancing the study of human-level color understanding of multimodal AI.

Contrastive Language-Image Pre-training (CLIP) has been a celebrated method for training vision encoders to generate image/text representations facilitating various applications. Recently, CLIP has been widely adopted as the vision backbone of multimodal large language models (MLLMs) to connect image inputs for language interactions. The success of CLIP as a vision-language foundation model relies on aligning web-crawled noisy text annotations at image levels. Nevertheless, such criteria may become insufficient for downstream tasks in need of fine-grained vision representations, especially when region-level understanding is demanding for MLLMs. In this paper, we improve the localization capability of CLIP with several advances. We propose a pre-training method called Contrastive Localized Language-Image Pre-training (CLOC) by complementing CLIP with region-text contrastive loss and modules. We formulate a new concept, promptable embeddings, of which the encoder produces image embeddings easy to transform into region representations given spatial hints. To support large-scale pre-training, we design a visually-enriched and spatially-localized captioning framework to effectively generate region-text pseudo-labels at scale. By scaling up to billions of annotated images, CLOC enables high-quality regional embeddings for image region recognition and retrieval tasks, and can be a drop-in replacement of CLIP to enhance MLLMs, especially on referring and grounding tasks.

Instruction tuning large language models (LLMs) using machine-generated instruction-following data has improved zero-shot capabilities on new tasks, but the idea is less explored in the multimodal field. In this paper, we present the first attempt to use language-only GPT-4 to generate multimodal language-image instruction-following data. By instruction tuning on such generated data, we introduce LLaVA: Large Language and Vision Assistant, an end-to-end trained large multimodal model that connects a vision encoder and LLM for general-purpose visual and language understanding.Our early experiments show that LLaVA demonstrates impressive multimodel chat abilities, sometimes exhibiting the behaviors of multimodal GPT-4 on unseen images/instructions, and yields a 85.1% relative score compared with GPT-4 on a synthetic multimodal instruction-following dataset. When fine-tuned on Science QA, the synergy of LLaVA and GPT-4 achieves a new state-of-the-art accuracy of 92.53%. We make GPT-4 generated visual instruction tuning data, our model and code base publicly available.

Vision Language Models (VLMs) like CLIP have attracted substantial attention in pathology, serving as backbones for applications such as zero-shot image classification and Whole Slide Image (WSI) analysis. Additionally, they can function as vision encoders when combined with large language models (LLMs) to support broader capabilities. Current efforts to train pathology VLMs rely on pathology image-text pairs from platforms like PubMed, YouTube, and Twitter, which provide limited, unscalable data with generally suboptimal image quality. In this work, we leverage large-scale WSI datasets like TCGA to extract numerous high-quality image patches. We then train a large multimodal model to generate captions for these images, creating PathGen-1.6M, a dataset containing 1.6 million high-quality image-caption pairs. Our approach involves multiple agent models collaborating to extract representative WSI patches, generating and refining captions to obtain high-quality image-text pairs. Extensive experiments show that integrating these generated pairs with existing datasets to train a pathology-specific CLIP model, PathGen-CLIP, significantly enhances its ability to analyze pathological images, with substantial improvements across nine pathology-related zero-shot image classification tasks and three whole-slide image tasks. Furthermore, we construct 200K instruction-tuning data based on PathGen-1.6M and integrate PathGen-CLIP with the Vicuna LLM to create more powerful multimodal models through instruction tuning. Overall, we provide a scalable pathway for high-quality data generation in pathology, paving the way for next-generation general pathology models.

Foundation models have emerged as powerful tools across various domains including language, vision, and multimodal tasks. While prior works have addressed unsupervised image segmentation, they significantly lag behind supervised models. In this paper, we use a diffusion UNet encoder as a foundation vision encoder and introduce DiffCut, an unsupervised zero-shot segmentation method that solely harnesses the output features from the final self-attention block. Through extensive experimentation, we demonstrate that the utilization of these diffusion features in a graph based segmentation algorithm, significantly outperforms previous state-of-the-art methods on zero-shot segmentation. Specifically, we leverage a recursive Normalized Cut algorithm that softly regulates the granularity of detected objects and produces well-defined segmentation maps that precisely capture intricate image details. Our work highlights the remarkably accurate semantic knowledge embedded within diffusion UNet encoders that could then serve as foundation vision encoders for downstream tasks. Project page at https://diffcut-segmentation.github.io

Recent studies indicate that large multimodal models (LMMs) are highly robust against natural distribution shifts, often surpassing previous baselines. Despite this, domain-specific adaptation is still necessary, particularly in specialized areas like healthcare. Due to the impracticality of fine-tuning LMMs given their vast parameter space, this work investigates in-context learning (ICL) as an effective alternative for enhancing LMMs' adaptability. We find that the success of ICL heavily relies on the choice of demonstration, mirroring challenges seen in large language models but introducing unique complexities for LMMs facing distribution shifts. Our study addresses this by evaluating an unsupervised ICL method, TopKNearestPR, which selects in-context examples through a nearest example search based on feature similarity. We uncover that its effectiveness is limited by the deficiencies of pre-trained vision encoders under distribution shift scenarios. To address these challenges, we propose InvariantSelectPR, a novel method leveraging Class-conditioned Contrastive Invariance (CCI) for more robust demonstration selection. Specifically, CCI enhances pre-trained vision encoders by improving their discriminative capabilities across different classes and ensuring invariance to domain-specific variations. This enhancement allows the encoders to effectively identify and retrieve the most informative examples, which are then used to guide LMMs in adapting to new query samples under varying distributions. Our experiments show that InvariantSelectPR substantially improves the adaptability of LMMs, achieving significant performance gains on benchmark datasets, with a 34.2%uparrow accuracy increase in 7-shot on Camelyon17 and 16.9%uparrow increase in 7-shot on HAM10000 compared to the baseline zero-shot performance.

Building general-purpose models that can effectively perceive the world through multimodal signals has been a long-standing goal. Current approaches involve integrating separately pre-trained components, such as connecting vision encoders to LLMs and continuing multimodal training. While such approaches exhibit remarkable sample efficiency, it remains an open question whether such late-fusion architectures are inherently superior. In this work, we revisit the architectural design of native multimodal models (NMMs)--those trained from the ground up on all modalities--and conduct an extensive scaling laws study, spanning 457 trained models with different architectures and training mixtures. Our investigation reveals no inherent advantage to late-fusion architectures over early-fusion ones, which do not rely on image encoders. On the contrary, early-fusion exhibits stronger performance at lower parameter counts, is more efficient to train, and is easier to deploy. Motivated by the strong performance of the early-fusion architectures, we show that incorporating Mixture of Experts (MoEs) allows for models that learn modality-specific weights, significantly enhancing performance.

Contrastively-trained Vision-Language Models (VLMs) like CLIP have become the de facto approach for discriminative vision-language representation learning. However, these models have limited language understanding, often exhibiting a "bag of words" behavior. At the same time, Large Vision-Language Models (LVLMs), which combine vision encoders with LLMs, have been shown capable of detailed vision-language reasoning, yet their autoregressive nature renders them less suitable for discriminative tasks. In this work, we propose to combine "the best of both worlds": a new training approach for discriminative fine-tuning of LVLMs that results in strong discriminative and compositional capabilities. Essentially, our approach converts a generative LVLM into a discriminative one, unlocking its capability for powerful image-text discrimination combined with enhanced language understanding. Our contributions include: (1) A carefully designed training/optimization framework that utilizes image-text pairs of variable length and granularity for training the model with both contrastive and next-token prediction losses. This is accompanied by ablation studies that justify the necessity of our framework's components. (2) A parameter-efficient adaptation method using a combination of soft prompting and LoRA adapters. (3) Significant improvements over state-of-the-art CLIP-like models of similar size, including standard image-text retrieval benchmarks and notable gains in compositionality.

Recent advancements in Multimodal Large Language Models (LLMs) have focused primarily on scaling by increasing text-image pair data and enhancing LLMs to improve performance on multimodal tasks. However, these scaling approaches are computationally expensive and overlook the significance of improving model capabilities from the vision side. Inspired by the successful applications of Mixture-of-Experts (MoE) in LLMs, which improves model scalability during training while keeping inference costs similar to those of smaller models, we propose CuMo. CuMo incorporates Co-upcycled Top-K sparsely-gated Mixture-of-experts blocks into both the vision encoder and the MLP connector, thereby enhancing the multimodal LLMs with minimal additional activated parameters during inference. CuMo first pre-trains the MLP blocks and then initializes each expert in the MoE block from the pre-trained MLP block during the visual instruction tuning stage. Auxiliary losses are used to ensure a balanced loading of experts. CuMo outperforms state-of-the-art multimodal LLMs across various VQA and visual-instruction-following benchmarks using models within each model size group, all while training exclusively on open-sourced datasets. The code and model weights for CuMo are open-sourced at https://github.com/SHI-Labs/CuMo.

Visuomotor robot policies, increasingly pre-trained on large-scale datasets, promise significant advancements across robotics domains. However, aligning these policies with end-user preferences remains a challenge, particularly when the preferences are hard to specify. While reinforcement learning from human feedback (RLHF) has become the predominant mechanism for alignment in non-embodied domains like large language models, it has not seen the same success in aligning visuomotor policies due to the prohibitive amount of human feedback required to learn visual reward functions. To address this limitation, we propose Representation-Aligned Preference-based Learning (RAPL), an observation-only method for learning visual rewards from significantly less human preference feedback. Unlike traditional RLHF, RAPL focuses human feedback on fine-tuning pre-trained vision encoders to align with the end-user's visual representation and then constructs a dense visual reward via feature matching in this aligned representation space. We first validate RAPL through simulation experiments in the X-Magical benchmark and Franka Panda robotic manipulation, demonstrating that it can learn rewards aligned with human preferences, more efficiently uses preference data, and generalizes across robot embodiments. Finally, our hardware experiments align pre-trained Diffusion Policies for three object manipulation tasks. We find that RAPL can fine-tune these policies with 5x less real human preference data, taking the first step towards minimizing human feedback while maximizing visuomotor robot policy alignment.

We introduce a radiology-focused visual language model designed to generate radiology reports from chest X-rays. Building on previous findings that large language models (LLMs) can acquire multimodal capabilities when aligned with pretrained vision encoders, we demonstrate similar potential with chest X-ray images. This integration enhances the ability of model to understand and describe chest X-ray images. Our model combines an image encoder with a fine-tuned LLM based on the Vicuna-7B architecture, enabling it to generate different sections of a radiology report with notable accuracy. The training process involves a two-stage approach: (i) initial alignment of chest X-ray features with the LLM (ii) followed by fine-tuning for radiology report generation.

We observe that the mapping between an image's representation in one model to its representation in another can be learned surprisingly well with just a linear layer, even across diverse models. Building on this observation, we propose text-to-concept, where features from a fixed pretrained model are aligned linearly to the CLIP space, so that text embeddings from CLIP's text encoder become directly comparable to the aligned features. With text-to-concept, we convert fixed off-the-shelf vision encoders to surprisingly strong zero-shot classifiers for free, with accuracy at times even surpassing that of CLIP, despite being much smaller models and trained on a small fraction of the data compared to CLIP. We show other immediate use-cases of text-to-concept, like building concept bottleneck models with no concept supervision, diagnosing distribution shifts in terms of human concepts, and retrieving images satisfying a set of text-based constraints. Lastly, we demonstrate the feasibility of concept-to-text, where vectors in a model's feature space are decoded by first aligning to the CLIP before being fed to a GPT-based generative model. Our work suggests existing deep models, with presumably diverse architectures and training, represent input samples relatively similarly, and a two-way communication across model representation spaces and to humans (through language) is viable.

Recent advancements in Multimodal Large Language Models (MLLMs) have revolutionized the field of vision-language understanding by integrating visual perception capabilities into Large Language Models (LLMs). The prevailing trend in this field involves the utilization of a vision encoder derived from vision-language contrastive learning (CL), showing expertise in capturing overall representations while facing difficulties in capturing detailed local patterns. In this work, we focus on enhancing the visual representations for MLLMs by combining high-frequency and detailed visual representations, obtained through masked image modeling (MIM), with semantically-enriched low-frequency representations captured by CL. To achieve this goal, we introduce X-Former which is a lightweight transformer module designed to exploit the complementary strengths of CL and MIM through an innovative interaction mechanism. Specifically, X-Former first bootstraps vision-language representation learning and multimodal-to-multimodal generative learning from two frozen vision encoders, i.e., CLIP-ViT (CL-based) and MAE-ViT (MIM-based). It further bootstraps vision-to-language generative learning from a frozen LLM to ensure visual features from X-Former can be interpreted by the LLM. To demonstrate the effectiveness of our approach, we assess its performance on tasks demanding detailed visual understanding. Extensive evaluations indicate that X-Former excels in visual reasoning tasks involving both structural and semantic categories in the GQA dataset. Assessment on fine-grained visual perception benchmark further confirms its superior capabilities in visual understanding.

Radiology report generation (RRG) is a challenging task, as it requires a thorough understanding of medical images, integration of multiple temporal inputs, and accurate report generation. Effective interpretation of medical images, such as chest X-rays (CXRs), demands sophisticated visual-language reasoning to map visual findings to structured reports. Recent studies have shown that multimodal large language models (MLLMs) can acquire multimodal capabilities by aligning with pre-trained vision encoders. However, current approaches predominantly focus on single-image analysis or utilise rule-based symbolic processing to handle multiple images, thereby overlooking the essential temporal information derived from comparing current images with prior ones. To overcome this critical limitation, we introduce Libra, a temporal-aware MLLM tailored for CXR report generation using temporal images. Libra integrates a radiology-specific image encoder with a MLLM and utilises a novel Temporal Alignment Connector to capture and synthesise temporal information of images across different time points with unprecedented precision. Extensive experiments show that Libra achieves new state-of-the-art performance among the same parameter scale MLLMs for RRG tasks on the MIMIC-CXR. Specifically, Libra improves the RadCliQ metric by 12.9% and makes substantial gains across all lexical metrics compared to previous models.

Business Document Information Extraction (BDIE) is the problem of transforming a blob of unstructured information (raw text, scanned documents, etc.) into a structured format that downstream systems can parse and use. It has two main tasks: Key-Information Extraction (KIE) and Line Items Recognition (LIR). In this paper, we argue that BDIE is best modeled as a Tool Use problem, where the tools are these downstream systems. We then present Retrieval Augmented Structured Generation (RASG), a novel general framework for BDIE that achieves state of the art (SOTA) results on both KIE and LIR tasks on BDIE benchmarks. The contributions of this paper are threefold: (1) We show, with ablation benchmarks, that Large Language Models (LLMs) with RASG are already competitive with or surpasses current SOTA Large Multimodal Models (LMMs) without RASG on BDIE benchmarks. (2) We propose a new metric class for Line Items Recognition, General Line Items Recognition Metric (GLIRM), that is more aligned with practical BDIE use cases compared to existing metrics, such as ANLS*, DocILE, and GriTS. (3) We provide a heuristic algorithm for backcalculating bounding boxes of predicted line items and tables without the need for vision encoders. Finally, we claim that, while LMMs might sometimes offer marginal performance benefits, LLMs + RASG is oftentimes superior given real-world applications and constraints of BDIE.

We present a radiology-specific multimodal model for the task for generating radiological reports from chest X-rays (CXRs). Our work builds on the idea that large language model(s) can be equipped with multimodal capabilities through alignment with pre-trained vision encoders. On natural images, this has been shown to allow multimodal models to gain image understanding and description capabilities. Our proposed model (MAIRA-1) leverages a CXR-specific image encoder in conjunction with a fine-tuned large language model based on Vicuna-7B, and text-based data augmentation, to produce reports with state-of-the-art quality. In particular, MAIRA-1 significantly improves on the radiologist-aligned RadCliQ metric and across all lexical metrics considered. Manual review of model outputs demonstrates promising fluency and accuracy of generated reports while uncovering failure modes not captured by existing evaluation practices. More information and resources can be found on the project website: https://aka.ms/maira.

Few-Shot Class Incremental Learning (FSCIL) is a challenging continual learning task, where limited training examples are available during several learning sessions. To succeed in this task, it is necessary to avoid over-fitting new classes caused by biased distributions in the few-shot training sets. The general approach to address this issue involves enhancing the representational capability of a pre-defined backbone architecture by adding special modules for backward compatibility with older classes. However, this approach has not yet solved the dilemma of ensuring high classification accuracy over time while reducing the gap between the performance obtained on larger training sets and the smaller ones. In this work, we propose an alternative approach called Continual Parameter-Efficient CLIP (CPE-CLIP) to reduce the loss of information between different learning sessions. Instead of adapting additional modules to address information loss, we leverage the vast knowledge acquired by CLIP in large-scale pre-training and its effectiveness in generalizing to new concepts. Our approach is multimodal and parameter-efficient, relying on learnable prompts for both the language and vision encoders to enable transfer learning across sessions. We also introduce prompt regularization to improve performance and prevent forgetting. Our experimental results demonstrate that CPE-CLIP significantly improves FSCIL performance compared to state-of-the-art proposals while also drastically reducing the number of learnable parameters and training costs.

In the realm of Multimodal Large Language Models (MLLMs), vision-language connector plays a crucial role to link the pre-trained vision encoders with Large Language Models (LLMs). Despite its importance, the vision-language connector has been relatively less explored. In this study, we aim to propose a strong vision-language connector that enables MLLMs to achieve high accuracy while maintain low computation cost. We first reveal the existence of the visual anchors in Vision Transformer and propose a cost-effective search algorithm to extract them. Building on these findings, we introduce the Anchor Former (AcFormer), a novel vision-language connector designed to leverage the rich prior knowledge obtained from these visual anchors during pretraining, guiding the aggregation of information. Through extensive experimentation, we demonstrate that the proposed method significantly reduces computational costs by nearly two-thirds compared with baseline, while simultaneously outperforming baseline methods. This highlights the effectiveness and efficiency of AcFormer.

Multimodal large language models (MLLMs) have attracted widespread interest and have rich applications. However, the inherent attention mechanism in its Transformer structure requires quadratic complexity and results in expensive computational overhead. Therefore, in this work, we propose VL-Mamba, a multimodal large language model based on state space models, which have been shown to have great potential for long-sequence modeling with fast inference and linear scaling in sequence length. Specifically, we first replace the transformer-based backbone language model such as LLama or Vicuna with the pre-trained Mamba language model. Then, we empirically explore how to effectively apply the 2D vision selective scan mechanism for multimodal learning and the combinations of different vision encoders and variants of pretrained Mamba language models. The extensive experiments on diverse multimodal benchmarks with competitive performance show the effectiveness of our proposed VL-Mamba and demonstrate the great potential of applying state space models for multimodal learning tasks.

Recent Multimodal Large Language Models (MLLMs) have achieved remarkable performance but face deployment challenges due to their quadratic computational complexity, growing Key-Value cache requirements, and reliance on separate vision encoders. We propose mmMamba, a framework for developing linear-complexity native multimodal state space models through progressive distillation from existing MLLMs using moderate academic computational resources. Our approach enables the direct conversion of trained decoder-only MLLMs to linear-complexity architectures without requiring pre-trained RNN-based LLM or vision encoders. We propose an seeding strategy to carve Mamba from trained Transformer and a three-stage distillation recipe, which can effectively transfer the knowledge from Transformer to Mamba while preserving multimodal capabilities. Our method also supports flexible hybrid architectures that combine Transformer and Mamba layers for customizable efficiency-performance trade-offs. Distilled from the Transformer-based decoder-only HoVLE, mmMamba-linear achieves competitive performance against existing linear and quadratic-complexity VLMs, while mmMamba-hybrid further improves performance significantly, approaching HoVLE's capabilities. At 103K tokens, mmMamba-linear demonstrates 20.6times speedup and 75.8% GPU memory reduction compared to HoVLE, while mmMamba-hybrid achieves 13.5times speedup and 60.2% memory savings. Code and models are released at https://github.com/hustvl/mmMamba

We introduce InternVL 2.5, an advanced multimodal large language model (MLLM) series that builds upon InternVL 2.0, maintaining its core model architecture while introducing significant enhancements in training and testing strategies as well as data quality. In this work, we delve into the relationship between model scaling and performance, systematically exploring the performance trends in vision encoders, language models, dataset sizes, and test-time configurations. Through extensive evaluations on a wide range of benchmarks, including multi-discipline reasoning, document understanding, multi-image / video understanding, real-world comprehension, multimodal hallucination detection, visual grounding, multilingual capabilities, and pure language processing, InternVL 2.5 exhibits competitive performance, rivaling leading commercial models such as GPT-4o and Claude-3.5-Sonnet. Notably, our model is the first open-source MLLMs to surpass 70% on the MMMU benchmark, achieving a 3.7-point improvement through Chain-of-Thought (CoT) reasoning and showcasing strong potential for test-time scaling. We hope this model contributes to the open-source community by setting new standards for developing and applying multimodal AI systems. HuggingFace demo see https://huggingface.co/spaces/OpenGVLab/InternVL

Despite the rapid integration of video perception capabilities into Large Multimodal Models (LMMs), the underlying mechanisms driving their video understanding remain poorly understood. Consequently, many design decisions in this domain are made without proper justification or analysis. The high computational cost of training and evaluating such models, coupled with limited open research, hinders the development of video-LMMs. To address this, we present a comprehensive study that helps uncover what effectively drives video understanding in LMMs. We begin by critically examining the primary contributors to the high computational requirements associated with video-LMM research and discover Scaling Consistency, wherein design and training decisions made on smaller models and datasets (up to a critical size) effectively transfer to larger models. Leveraging these insights, we explored many video-specific aspects of video-LMMs, including video sampling, architectures, data composition, training schedules, and more. For example, we demonstrated that fps sampling during training is vastly preferable to uniform frame sampling and which vision encoders are the best for video representation. Guided by these findings, we introduce Apollo, a state-of-the-art family of LMMs that achieve superior performance across different model sizes. Our models can perceive hour-long videos efficiently, with Apollo-3B outperforming most existing 7B models with an impressive 55.1 on LongVideoBench. Apollo-7B is state-of-the-art compared to 7B LMMs with a 70.9 on MLVU, and 63.3 on Video-MME.

In Multimodal Large Language Models (MLLMs), a visual projector plays a crucial role in bridging pre-trained vision encoders with LLMs, enabling profound visual understanding while harnessing the LLMs' robust capabilities. Despite the importance of the visual projector, it has been relatively less explored. In this study, we first identify two essential projector properties: (i) flexibility in managing the number of visual tokens, crucial for MLLMs' overall efficiency, and (ii) preservation of local context from visual features, vital for spatial understanding. Based on these findings, we propose a novel projector design that is both flexible and locality-enhanced, effectively satisfying the two desirable properties. Additionally, we present comprehensive strategies to effectively utilize multiple and multifaceted instruction datasets. Through extensive experiments, we examine the impact of individual design choices. Finally, our proposed MLLM, Honeybee, remarkably outperforms previous state-of-the-art methods across various benchmarks, including MME, MMBench, SEED-Bench, and LLaVA-Bench, achieving significantly higher efficiency. Code and models are available at https://github.com/kakaobrain/honeybee.

Chain-of-Thought (CoT) has widely enhanced mathematical reasoning in Large Language Models (LLMs), but it still remains challenging for extending it to multimodal domains. Existing works either adopt a similar textual reasoning for image input, or seek to interleave visual signals into mathematical CoT. However, they face three key limitations for math problem-solving: reliance on coarse-grained box-shaped image regions, limited perception of vision encoders on math content, and dependence on external capabilities for visual modification. In this paper, we propose MINT-CoT, introducing Mathematical INterleaved Tokens for Chain-of-Thought visual reasoning. MINT-CoT adaptively interleaves relevant visual tokens into textual reasoning steps via an Interleave Token, which dynamically selects visual regions of any shapes within math figures. To empower this capability, we construct the MINT-CoT dataset, containing 54K mathematical problems aligning each reasoning step with visual regions at the token level, accompanied by a rigorous data generation pipeline. We further present a three-stage MINT-CoT training strategy, progressively combining text-only CoT SFT, interleaved CoT SFT, and interleaved CoT RL, which derives our MINT-CoT-7B model. Extensive experiments demonstrate the effectiveness of our method for effective visual interleaved reasoning in mathematical domains, where MINT-CoT-7B outperforms the baseline model by +34.08% on MathVista, +28.78% on GeoQA, and +23.2% on MMStar, respectively. Our code and data are available at https://github.com/xinyan-cxy/MINT-CoT

The differing representation spaces required for visual understanding and generation pose a challenge in unifying them within the autoregressive paradigm of large language models. A vision tokenizer trained for reconstruction excels at capturing low-level perceptual details, making it well-suited for visual generation but lacking high-level semantic representations for understanding tasks. Conversely, a vision encoder trained via contrastive learning aligns well with language but struggles to decode back into the pixel space for generation tasks. To bridge this gap, we propose DualToken, a method that unifies representations for both understanding and generation within a single tokenizer. However, directly integrating reconstruction and semantic objectives in a single tokenizer creates conflicts, leading to degraded performance in both reconstruction quality and semantic performance. Instead of forcing a single codebook to handle both semantic and perceptual information, DualToken disentangles them by introducing separate codebooks for high and low-level features, effectively transforming their inherent conflict into a synergistic relationship. As a result, DualToken achieves state-of-the-art performance in both reconstruction and semantic tasks while demonstrating remarkable effectiveness in downstream MLLM understanding and generation tasks. Notably, we also show that DualToken, as a unified tokenizer, surpasses the naive combination of two distinct types vision encoders, providing superior performance within a unified MLLM.

The Bavarian Academy of Sciences and Humanities aims to digitize its Medieval Latin Dictionary. This dictionary entails record cards referring to lemmas in medieval Latin, a low-resource language. A crucial step of the digitization process is the Handwritten Text Recognition (HTR) of the handwritten lemmas found on these record cards. In our work, we introduce an end-to-end pipeline, tailored to the medieval Latin dictionary, for locating, extracting, and transcribing the lemmas. We employ two state-of-the-art (SOTA) image segmentation models to prepare the initial data set for the HTR task. Furthermore, we experiment with different transformer-based models and conduct a set of experiments to explore the capabilities of different combinations of vision encoders with a GPT-2 decoder. Additionally, we also apply extensive data augmentation resulting in a highly competitive model. The best-performing setup achieved a Character Error Rate (CER) of 0.015, which is even superior to the commercial Google Cloud Vision model, and shows more stable performance.

We introduce new jailbreak attacks on vision language models (VLMs), which use aligned LLMs and are resilient to text-only jailbreak attacks. Specifically, we develop cross-modality attacks on alignment where we pair adversarial images going through the vision encoder with textual prompts to break the alignment of the language model. Our attacks employ a novel compositional strategy that combines an image, adversarially targeted towards toxic embeddings, with generic prompts to accomplish the jailbreak. Thus, the LLM draws the context to answer the generic prompt from the adversarial image. The generation of benign-appearing adversarial images leverages a novel embedding-space-based methodology, operating with no access to the LLM model. Instead, the attacks require access only to the vision encoder and utilize one of our four embedding space targeting strategies. By not requiring access to the LLM, the attacks lower the entry barrier for attackers, particularly when vision encoders such as CLIP are embedded in closed-source LLMs. The attacks achieve a high success rate across different VLMs, highlighting the risk of cross-modality alignment vulnerabilities, and the need for new alignment approaches for multi-modal models.

Stimulated by the sophisticated reasoning capabilities of recent Large Language Models (LLMs), a variety of strategies for bridging video modality have been devised. A prominent strategy involves Video Language Models (VideoLMs), which train a learnable interface with video data to connect advanced vision encoders with LLMs. Recently, an alternative strategy has surfaced, employing readily available foundation models, such as VideoLMs and LLMs, across multiple stages for modality bridging. In this study, we introduce a simple yet novel strategy where only a single Vision Language Model (VLM) is utilized. Our starting point is the plain insight that a video comprises a series of images, or frames, interwoven with temporal information. The essence of video comprehension lies in adeptly managing the temporal aspects along with the spatial details of each frame. Initially, we transform a video into a single composite image by arranging multiple frames in a grid layout. The resulting single image is termed as an image grid. This format, while maintaining the appearance of a solitary image, effectively retains temporal information within the grid structure. Therefore, the image grid approach enables direct application of a single high-performance VLM without necessitating any video-data training. Our extensive experimental analysis across ten zero-shot video question answering benchmarks, including five open-ended and five multiple-choice benchmarks, reveals that the proposed Image Grid Vision Language Model (IG-VLM) surpasses the existing methods in nine out of ten benchmarks.

Recent progress in unified models for image understanding and generation has been impressive, yet most approaches remain limited to single-modal generation conditioned on multiple modalities. In this paper, we present Mogao, a unified framework that advances this paradigm by enabling interleaved multi-modal generation through a causal approach. Mogao integrates a set of key technical improvements in architecture design, including a deep-fusion design, dual vision encoders, interleaved rotary position embeddings, and multi-modal classifier-free guidance, which allow it to harness the strengths of both autoregressive models for text generation and diffusion models for high-quality image synthesis. These practical improvements also make Mogao particularly effective to process interleaved sequences of text and images arbitrarily. To further unlock the potential of unified models, we introduce an efficient training strategy on a large-scale, in-house dataset specifically curated for joint text and image generation. Extensive experiments show that Mogao not only achieves state-of-the-art performance in multi-modal understanding and text-to-image generation, but also excels in producing high-quality, coherent interleaved outputs. Its emergent capabilities in zero-shot image editing and compositional generation highlight Mogao as a practical omni-modal foundation model, paving the way for future development and scaling the unified multi-modal systems.

Transductive zero-shot learning with vision-language models leverages image-image similarities within the dataset to achieve better classification accuracy compared to the inductive setting. However, there is little work that explores the structure of the language space in this context. We propose GTA-CLIP, a novel technique that incorporates supervision from language models for joint transduction in language and vision spaces. Our approach is iterative and consists of three steps: (i) incrementally exploring the attribute space by querying language models, (ii) an attribute-augmented transductive inference procedure, and (iii) fine-tuning the language and vision encoders based on inferred labels within the dataset. Through experiments with CLIP encoders, we demonstrate that GTA-CLIP, yields an average performance improvement of 8.6% and 3.7% across 12 datasets and 3 encoders, over CLIP and transductive CLIP respectively in the zero-shot setting. We also observe similar improvements in a few-shot setting. We present ablation studies that demonstrate the value of each step and visualize how the vision and language spaces evolve over iterations driven by the transductive learning.

The goal of object-centric representation learning is to decompose visual scenes into a structured representation that isolates the entities. Recent successes have shown that object-centric representation learning can be scaled to real-world scenes by utilizing pre-trained self-supervised features. However, so far, object-centric methods have mostly been applied in-distribution, with models trained and evaluated on the same dataset. This is in contrast to the wider trend in machine learning towards general-purpose models directly applicable to unseen data and tasks. Thus, in this work, we study current object-centric methods through the lens of zero-shot generalization by introducing a benchmark comprising eight different synthetic and real-world datasets. We analyze the factors influencing zero-shot performance and find that training on diverse real-world images improves transferability to unseen scenarios. Furthermore, inspired by the success of task-specific fine-tuning in foundation models, we introduce a novel fine-tuning strategy to adapt pre-trained vision encoders for the task of object discovery. We find that the proposed approach results in state-of-the-art performance for unsupervised object discovery, exhibiting strong zero-shot transfer to unseen datasets.

Large Language Models (LLMs) have demonstrated exceptional text understanding. Existing works explore their application in text embedding tasks. However, there are few works utilizing LLMs to assist multimodal representation tasks. In this work, we investigate the potential of LLMs to enhance multimodal representation in multimodal item-to-item (I2I) recommendations. One feasible method is the transfer of Multimodal Large Language Models (MLLMs) for representation tasks. However, pre-training MLLMs usually requires collecting high-quality, web-scale multimodal data, resulting in complex training procedures and high costs. This leads the community to rely heavily on open-source MLLMs, hindering customized training for representation scenarios. Therefore, we aim to design an end-to-end training method that customizes the integration of any existing LLMs and vision encoders to construct efficient multimodal representation models. Preliminary experiments show that fine-tuned LLMs in this end-to-end method tend to overlook image content. To overcome this challenge, we propose a novel training framework, NoteLLM-2, specifically designed for multimodal representation. We propose two ways to enhance the focus on visual information. The first method is based on the prompt viewpoint, which separates multimodal content into visual content and textual content. NoteLLM-2 adopts the multimodal In-Content Learning method to teach LLMs to focus on both modalities and aggregate key information. The second method is from the model architecture, utilizing a late fusion mechanism to directly fuse visual information into textual information. Extensive experiments have been conducted to validate the effectiveness of our method.

Recent document question answering models consist of two key components: the vision encoder, which captures layout and visual elements in images, and a Large Language Model (LLM) that helps contextualize questions to the image and supplements them with external world knowledge to generate accurate answers. However, the relative contributions of the vision encoder and the language model in these tasks remain unclear. This is especially interesting given the effectiveness of instruction-tuned LLMs, which exhibit remarkable adaptability to new tasks. To this end, we explore the following aspects in this work: (1) The efficacy of an LLM-only approach on document question answering tasks (2) strategies for serializing textual information within document images and feeding it directly to an instruction-tuned LLM, thus bypassing the need for an explicit vision encoder (3) thorough quantitative analysis on the feasibility of such an approach. Our comprehensive analysis encompasses six diverse benchmark datasets, utilizing LLMs of varying scales. Our findings reveal that a strategy exclusively reliant on the LLM yields results that are on par with or closely approach state-of-the-art performance across a range of datasets. We posit that this evaluation framework will serve as a guiding resource for selecting appropriate datasets for future research endeavors that emphasize the fundamental importance of layout and image content information.

Federated Learning (FL) has gained popularity for fine-tuning large language models (LLMs) across multiple nodes, each with its own private data. While LoRA has been widely adopted for parameter efficient federated fine-tuning, recent theoretical and empirical studies highlight its suboptimal performance in the federated learning context. In response, we propose a novel, simple, and more effective parameter-efficient fine-tuning method that does not rely on LoRA. Our approach introduces a small multi-layer perceptron (MLP) layer between two existing MLP layers the up proj (the FFN projection layer following the self-attention module) and down proj within the feed forward network of the transformer block. This solution addresses the bottlenecks associated with LoRA in federated fine tuning and outperforms recent LoRA-based approaches, demonstrating superior performance for both language models and vision encoders.

Procedural video representation learning is an active research area where the objective is to learn an agent which can anticipate and forecast the future given the present video input, typically in conjunction with textual annotations. Prior works often rely on large-scale pretraining of visual encoders and prediction models with language supervision. However, the necessity and effectiveness of extending compute intensive pretraining to learn video clip sequences with noisy text supervision have not yet been fully validated by previous works. In this work, we show that a strong off-the-shelf frozen pretrained visual encoder, along with a well designed prediction model, can achieve state-of-the-art (SoTA) performance in forecasting and procedural planning without the need for pretraining the prediction model, nor requiring additional supervision from language or ASR. Instead of learning representations from pixel space, our method utilizes the latent embedding space of publicly available vision encoders. By conditioning on frozen clip-level embeddings from observed steps to predict the actions of unseen steps, our prediction model is able to learn robust representations for forecasting through iterative denoising - leveraging the recent advances in diffusion transformers (Peebles & Xie, 2023). Empirical studies over a total of five procedural learning tasks across four datasets (NIV, CrossTask, COIN and Ego4D-v2) show that our model advances the strong baselines in long-horizon action anticipation (+2.6% in Verb ED@20, +3.1% in Noun ED@20), and significantly improves the SoTA in step forecasting (+5.0%), task classification (+3.8%), and procedure planning tasks (up to +2.28% in success rate, +3.39% in mAcc, and +0.90% in mIoU).

Large language models (LLMs) excel in various tasks but are primarily trained on text data, limiting their application scope. Expanding LLM capabilities to include vision-language understanding is vital, yet training them on multimodal data from scratch is challenging and costly. Existing instruction tuning methods, e.g., LLAVA, often connects a pretrained CLIP vision encoder and LLMs via fully fine-tuning LLMs to bridge the modality gap. However, full fine-tuning is plagued by catastrophic forgetting, i.e., forgetting previous knowledge, and high training costs particularly in the era of increasing tasks and modalities. To solve this issue, we introduce MoExtend, an effective framework designed to streamline the modality adaptation and extension of Mixture-of-Experts (MoE) models. MoExtend seamlessly integrates new experts into pre-trained MoE models, endowing them with novel knowledge without the need to tune pretrained models such as MoE and vision encoders. This approach enables rapid adaptation and extension to new modal data or tasks, effectively addressing the challenge of accommodating new modalities within LLMs. Furthermore, MoExtend avoids tuning pretrained models, thus mitigating the risk of catastrophic forgetting. Experimental results demonstrate the efficacy and efficiency of MoExtend in enhancing the multimodal capabilities of LLMs, contributing to advancements in multimodal AI research. Code: https://github.com/zhongshsh/MoExtend.

Visual Self-Supervised Learning (SSL) currently underperforms Contrastive Language-Image Pretraining (CLIP) in multimodal settings such as Visual Question Answering (VQA). This multimodal gap is often attributed to the semantics introduced by language supervision, even though visual SSL and CLIP models are often trained on different data. In this work, we ask the question: "Do visual self-supervised approaches lag behind CLIP due to the lack of language supervision, or differences in the training data?" We study this question by training both visual SSL and CLIP models on the same MetaCLIP data, and leveraging VQA as a diverse testbed for vision encoders. In this controlled setup, visual SSL models scale better than CLIP models in terms of data and model capacity, and visual SSL performance does not saturate even after scaling up to 7B parameters. Consequently, we observe visual SSL methods achieve CLIP-level performance on a wide range of VQA and classic vision benchmarks. These findings demonstrate that pure visual SSL can match language-supervised visual pretraining at scale, opening new opportunities for vision-centric representation learning.

We introduce SigLIP 2, a family of new multilingual vision-language encoders that build on the success of the original SigLIP. In this second iteration, we extend the original image-text training objective with several prior, independently developed techniques into a unified recipe -- this includes captioning-based pretraining, self-supervised losses (self-distillation, masked prediction) and online data curation. With these changes, SigLIP 2 models outperform their SigLIP counterparts at all model scales in core capabilities, including zero-shot classification, image-text retrieval, and transfer performance when extracting visual representations for Vision-Language Models (VLMs). Furthermore, the new training recipe leads to significant improvements on localization and dense prediction tasks. We also train variants which support multiple resolutions and preserve the input's native aspect ratio. Finally, we train on a more diverse data-mixture that includes de-biasing techniques, leading to much better multilingual understanding and improved fairness. To allow users to trade off inference cost with performance, we release model checkpoints at four sizes: ViT-B (86M), L (303M), So400m (400M), and g (1B).

In the past several years there has been an explosion of available models for vision-language tasks. Unfortunately, the literature still leaves open a number of questions related to best practices in designing and training such models. In this paper we seek to answer several questions related to the pretraining of vision-language encoders through meta-analysis. In our first set of experiments, we show that we can save significant compute at no cost to downstream performance, by freezing large parts of vision-language models during pretraining. In our second set of experiments we examine the effect of basing a VL transformer on a vision model versus a text model. Additionally, we introduce a VL modeling platform called Renaissance that we use to conduct all of the experiments. This program offers a great deal of flexibility in creating, training and evaluating transformer encoders for VL modeling. The source code for Renaissance can be found at https://github.com/bsu-slim/renaissance.

Vision-Language-Action (VLA) models aim to predict robotic actions based on visual observations and language instructions. Existing approaches require fine-tuning pre-trained visionlanguage models (VLMs) as visual and language features are independently fed into downstream policies, degrading the pre-trained semantic alignments. We propose OTTER, a novel VLA architecture that leverages these existing alignments through explicit, text-aware visual feature extraction. Instead of processing all visual features, OTTER selectively extracts and passes only task-relevant visual features that are semantically aligned with the language instruction to the policy transformer. This allows OTTER to keep the pre-trained vision-language encoders frozen. Thereby, OTTER preserves and utilizes the rich semantic understanding learned from large-scale pre-training, enabling strong zero-shot generalization capabilities. In simulation and real-world experiments, OTTER significantly outperforms existing VLA models, demonstrating strong zeroshot generalization to novel objects and environments. Video, code, checkpoints, and dataset: https://ottervla.github.io/.

Complex 3D scene understanding has gained increasing attention, with scene encoding strategies playing a crucial role in this success. However, the optimal scene encoding strategies for various scenarios remain unclear, particularly compared to their image-based counterparts. To address this issue, we present a comprehensive study that probes various visual encoding models for 3D scene understanding, identifying the strengths and limitations of each model across different scenarios. Our evaluation spans seven vision foundation encoders, including image-based, video-based, and 3D foundation models. We evaluate these models in four tasks: Vision-Language Scene Reasoning, Visual Grounding, Segmentation, and Registration, each focusing on different aspects of scene understanding. Our evaluations yield key findings: DINOv2 demonstrates superior performance, video models excel in object-level tasks, diffusion models benefit geometric tasks, and language-pretrained models show unexpected limitations in language-related tasks. These insights challenge some conventional understandings, provide novel perspectives on leveraging visual foundation models, and highlight the need for more flexible encoder selection in future vision-language and scene-understanding tasks.

Vision-Language (VL) models with the Two-Tower architecture have dominated visual-language representation learning in recent years. Current VL models either use lightweight uni-modal encoders and learn to extract, align and fuse both modalities simultaneously in a deep cross-modal encoder, or feed the last-layer uni-modal representations from the deep pre-trained uni-modal encoders into the top cross-modal encoder. Both approaches potentially restrict vision-language representation learning and limit model performance. In this paper, we propose BridgeTower, which introduces multiple bridge layers that build a connection between the top layers of uni-modal encoders and each layer of the cross-modal encoder. This enables effective bottom-up cross-modal alignment and fusion between visual and textual representations of different semantic levels of pre-trained uni-modal encoders in the cross-modal encoder. Pre-trained with only 4M images, BridgeTower achieves state-of-the-art performance on various downstream vision-language tasks. In particular, on the VQAv2 test-std set, BridgeTower achieves an accuracy of 78.73%, outperforming the previous state-of-the-art model METER by 1.09% with the same pre-training data and almost negligible additional parameters and computational costs. Notably, when further scaling the model, BridgeTower achieves an accuracy of 81.15%, surpassing models that are pre-trained on orders-of-magnitude larger datasets. Code and checkpoints are available at https://github.com/microsoft/BridgeTower.

Aligned text-image encoders such as CLIP have become the de facto model for vision-language tasks. Furthermore, modality-specific encoders achieve impressive performances in their respective domains. This raises a central question: does an alignment exist between uni-modal vision and language encoders since they fundamentally represent the same physical world? Analyzing the latent spaces structure of vision and language models on image-caption benchmarks using the Centered Kernel Alignment (CKA), we find that the representation spaces of unaligned and aligned encoders are semantically similar. In the absence of statistical similarity in aligned encoders like CLIP, we show that a possible matching of unaligned encoders exists without any training. We frame this as a seeded graph-matching problem exploiting the semantic similarity between graphs and propose two methods - a Fast Quadratic Assignment Problem optimization, and a novel localized CKA metric-based matching/retrieval. We demonstrate the effectiveness of this on several downstream tasks including cross-lingual, cross-domain caption matching and image classification. Code available at github.com/mayug/0-shot-llm-vision.

Parameter Efficient Tuning (PET) has gained attention for reducing the number of parameters while maintaining performance and providing better hardware resource savings, but few studies investigate dense prediction tasks and interaction between modalities. In this paper, we do an investigation of efficient tuning problems on referring image segmentation. We propose a novel adapter called Bridger to facilitate cross-modal information exchange and inject task-specific information into the pre-trained model. We also design a lightweight decoder for image segmentation. Our approach achieves comparable or superior performance with only 1.61\% to 3.38\% backbone parameter updates, evaluated on challenging benchmarks. The code is available at https://github.com/kkakkkka/ETRIS.

We release MiDaS v3.1 for monocular depth estimation, offering a variety of new models based on different encoder backbones. This release is motivated by the success of transformers in computer vision, with a large variety of pretrained vision transformers now available. We explore how using the most promising vision transformers as image encoders impacts depth estimation quality and runtime of the MiDaS architecture. Our investigation also includes recent convolutional approaches that achieve comparable quality to vision transformers in image classification tasks. While the previous release MiDaS v3.0 solely leverages the vanilla vision transformer ViT, MiDaS v3.1 offers additional models based on BEiT, Swin, SwinV2, Next-ViT and LeViT. These models offer different performance-runtime tradeoffs. The best model improves the depth estimation quality by 28% while efficient models enable downstream tasks requiring high frame rates. We also describe the general process for integrating new backbones. A video summarizing the work can be found at https://youtu.be/UjaeNNFf9sE and the code is available at https://github.com/isl-org/MiDaS.

Advances in Large Language Models (LLMs) have inspired a surge of research exploring their expansion into the visual domain. While recent models exhibit promise in generating abstract captions for images and conducting natural conversations, their performance on text-rich images leaves room for improvement. In this paper, we propose the Contrastive Reading Model (Cream), a novel neural architecture designed to enhance the language-image understanding capability of LLMs by capturing intricate details typically overlooked by existing methods. Cream integrates vision and auxiliary encoders, complemented by a contrastive feature alignment technique, resulting in a more effective understanding of textual information within document images. Our approach, thus, seeks to bridge the gap between vision and language understanding, paving the way for more sophisticated Document Intelligence Assistants. Rigorous evaluations across diverse tasks, such as visual question answering on document images, demonstrate the efficacy of Cream as a state-of-the-art model in the field of visual document understanding. We provide our codebase and newly-generated datasets at https://github.com/naver-ai/cream

In the domain of computer vision, the restoration of missing information in video frames is a critical challenge, particularly in applications such as autonomous driving and surveillance systems. This paper introduces the Siamese Masked Conditional Variational Autoencoder (SiamMCVAE), leveraging a siamese architecture with twin encoders based on vision transformers. This innovative design enhances the model's ability to comprehend lost content by capturing intrinsic similarities between paired frames. SiamMCVAE proficiently reconstructs missing elements in masked frames, effectively addressing issues arising from camera malfunctions through variational inferences. Experimental results robustly demonstrate the model's effectiveness in restoring missing information, thus enhancing the resilience of computer vision systems. The incorporation of Siamese Vision Transformer (SiamViT) encoders in SiamMCVAE exemplifies promising potential for addressing real-world challenges in computer vision, reinforcing the adaptability of autonomous systems in dynamic environments.

We study the task of language instruction-guided robotic manipulation, in which an embodied robot is supposed to manipulate the target objects based on the language instructions. In previous studies, the predicted manipulation regions of the target object typically do not change with specification from the language instructions, which means that the language perception and manipulation prediction are separate. However, in human behavioral patterns, the manipulation regions of the same object will change for different language instructions. In this paper, we propose Instruction-Guided Affordance Net (IGANet) for predicting affordance maps of instruction-guided robotic manipulation tasks by utilizing powerful priors from vision and language encoders pre-trained on large-scale datasets. We develop a Vison-Language-Models(VLMs)-based data augmentation pipeline, which can generate a large amount of data automatically for model training. Besides, with the help of Large-Language-Models(LLMs), actions can be effectively executed to finish the tasks defined by instructions. A series of real-world experiments revealed that our method can achieve better performance with generated data. Moreover, our model can generalize better to scenarios with unseen objects and language instructions.

This paper investigates robot manipulation based on human instruction with ambiguous requests. The intent is to compensate for imperfect natural language via visual observations. Early symbolic methods, based on manually defined symbols, built modular framework consist of semantic parsing and task planning for producing sequences of actions from natural language requests. Modern connectionist methods employ deep neural networks to automatically learn visual and linguistic features and map to a sequence of low-level actions, in an endto-end fashion. These two approaches are blended to create a hybrid, modular framework: it formulates instruction following as symbolic goal learning via deep neural networks followed by task planning via symbolic planners. Connectionist and symbolic modules are bridged with Planning Domain Definition Language. The vision-and-language learning network predicts its goal representation, which is sent to a planner for producing a task-completing action sequence. For improving the flexibility of natural language, we further incorporate implicit human intents with explicit human instructions. To learn generic features for vision and language, we propose to separately pretrain vision and language encoders on scene graph parsing and semantic textual similarity tasks. Benchmarking evaluates the impacts of different components of, or options for, the vision-and-language learning model and shows the effectiveness of pretraining strategies. Manipulation experiments conducted in the simulator AI2THOR show the robustness of the framework to novel scenarios.

Many high-level skills that are required for computer vision tasks, such as parsing questions, comparing and contrasting semantics, and writing descriptions, are also required in other domains such as natural language processing. In this paper, we ask whether it is possible to learn those skills from text data and then transfer them to vision tasks without ever training on visual training data. Key to our approach is exploiting the joint embedding space of contrastively trained vision and language encoders. In practice, there can be systematic differences between embedding spaces for different modalities in contrastive models, and we analyze how these differences affect our approach and study strategies to mitigate this concern. We produce models using only text training data on four representative tasks: image captioning, visual entailment, visual question answering and visual news captioning, and evaluate them on standard benchmarks using images. We find these models perform close to models trained on images, while surpassing prior work for captioning and visual entailment in this text-only setting by over 9 points, and outperforming all prior work on visual news by over 30 points. We also showcase a variety of stylistic image captioning models that are trained using no image data and no human-curated language data, but instead using readily-available text data from books, the web, or language models.

Contrastive Language-Image Pre-training (CLIP) models have shown significant potential, particularly in zero-shot classification across diverse distribution shifts. Building on existing evaluations of overall classification robustness, this work aims to provide a more comprehensive assessment of CLIP by introducing several new perspectives. First, we investigate their robustness to variations in specific visual factors. Second, we assess two critical safety objectives--confidence uncertainty and out-of-distribution detection--beyond mere classification accuracy. Third, we evaluate the finesse with which CLIP models bridge the image and text modalities. Fourth, we extend our examination to 3D awareness in CLIP models, moving beyond traditional 2D image understanding. Finally, we explore the interaction between vision and language encoders within modern large multimodal models (LMMs) that utilize CLIP as the visual backbone, focusing on how this interaction impacts classification robustness. In each aspect, we consider the impact of six factors on CLIP models: model architecture, training distribution, training set size, fine-tuning, contrastive loss, and test-time prompts. Our study uncovers several previously unknown insights into CLIP. For instance, the architecture of the visual encoder in CLIP plays a significant role in their robustness against 3D corruption. CLIP models tend to exhibit a bias towards shape when making predictions. Moreover, this bias tends to diminish after fine-tuning on ImageNet. Vision-language models like LLaVA, leveraging the CLIP vision encoder, could exhibit benefits in classification performance for challenging categories over CLIP alone. Our findings are poised to offer valuable guidance for enhancing the robustness and reliability of CLIP models.

Chest X-rays (CXRs) are the most frequently performed imaging test in clinical practice. Recent advances in the development of vision-language foundation models (FMs) give rise to the possibility of performing automated CXR interpretation, which can assist physicians with clinical decision-making and improve patient outcomes. However, developing FMs that can accurately interpret CXRs is challenging due to the (1) limited availability of large-scale vision-language datasets in the medical image domain, (2) lack of vision and language encoders that can capture the complexities of medical data, and (3) absence of evaluation frameworks for benchmarking the abilities of FMs on CXR interpretation. In this work, we address these challenges by first introducing CheXinstruct - a large-scale instruction-tuning dataset curated from 28 publicly-available datasets. We then present CheXagent - an instruction-tuned FM capable of analyzing and summarizing CXRs. To build CheXagent, we design a clinical large language model (LLM) for parsing radiology reports, a vision encoder for representing CXR images, and a network to bridge the vision and language modalities. Finally, we introduce CheXbench - a novel benchmark designed to systematically evaluate FMs across 8 clinically-relevant CXR interpretation tasks. Extensive quantitative evaluations and qualitative reviews with five expert radiologists demonstrate that CheXagent outperforms previously-developed general- and medical-domain FMs on CheXbench tasks. Furthermore, in an effort to improve model transparency, we perform a fairness evaluation across factors of sex, race and age to highlight potential performance disparities. Our project is at https://stanford-aimi.github.io/chexagent.html.

Recent works have shown that, when trained at scale, uni-modal 2D vision and text encoders converge to learned features that share remarkable structural properties, despite arising from different representations. However, the role of 3D encoders with respect to other modalities remains unexplored. Furthermore, existing 3D foundation models that leverage large datasets are typically trained with explicit alignment objectives with respect to frozen encoders from other representations. In this work, we investigate the possibility of a posteriori alignment of representations obtained from uni-modal 3D encoders compared to text-based feature spaces. We show that naive post-training feature alignment of uni-modal text and 3D encoders results in limited performance. We then focus on extracting subspaces of the corresponding feature spaces and discover that by projecting learned representations onto well-chosen lower-dimensional subspaces the quality of alignment becomes significantly higher, leading to improved accuracy on matching and retrieval tasks. Our analysis further sheds light on the nature of these shared subspaces, which roughly separate between semantic and geometric data representations. Overall, ours is the first work that helps to establish a baseline for post-training alignment of 3D uni-modal and text feature spaces, and helps to highlight both the shared and unique properties of 3D data compared to other representations.

We present an unsupervised visual feature learning algorithm driven by context-based pixel prediction. By analogy with auto-encoders, we propose Context Encoders -- a convolutional neural network trained to generate the contents of an arbitrary image region conditioned on its surroundings. In order to succeed at this task, context encoders need to both understand the content of the entire image, as well as produce a plausible hypothesis for the missing part(s). When training context encoders, we have experimented with both a standard pixel-wise reconstruction loss, as well as a reconstruction plus an adversarial loss. The latter produces much sharper results because it can better handle multiple modes in the output. We found that a context encoder learns a representation that captures not just appearance but also the semantics of visual structures. We quantitatively demonstrate the effectiveness of our learned features for CNN pre-training on classification, detection, and segmentation tasks. Furthermore, context encoders can be used for semantic inpainting tasks, either stand-alone or as initialization for non-parametric methods.

Currently, vision encoder models like Vision Transformers (ViTs) typically excel at image recognition tasks but cannot simultaneously support text recognition like human visual recognition. To address this limitation, we propose UNIT, a novel training framework aimed at UNifying Image and Text recognition within a single model. Starting with a vision encoder pre-trained with image recognition tasks, UNIT introduces a lightweight language decoder for predicting text outputs and a lightweight vision decoder to prevent catastrophic forgetting of the original image encoding capabilities. The training process comprises two stages: intra-scale pretraining and inter-scale finetuning. During intra-scale pretraining, UNIT learns unified representations from multi-scale inputs, where images and documents are at their commonly used resolution, to enable fundamental recognition capability. In the inter-scale finetuning stage, the model introduces scale-exchanged data, featuring images and documents at resolutions different from the most commonly used ones, to enhance its scale robustness. Notably, UNIT retains the original vision encoder architecture, making it cost-free in terms of inference and deployment. Experiments across multiple benchmarks confirm that our method significantly outperforms existing methods on document-related tasks (e.g., OCR and DocQA) while maintaining the performances on natural images, demonstrating its ability to substantially enhance text recognition without compromising its core image recognition capabilities.

In recent years, multimodal large language models (MLLMs) have shown strong potential in real-world applications. They are developing rapidly due to their remarkable ability to comprehend multimodal information and their inherent powerful cognitive and reasoning capabilities. Among MLLMs, vision language models (VLM) stand out for their ability to understand vision information. However, the scaling trend of VLMs under the current mainstream paradigm has not been extensively studied. Whether we can achieve better performance by training even larger models is still unclear. To address this issue, we conducted experiments on the pretraining stage of MLLMs. We conduct our experiment using different encoder sizes and large language model (LLM) sizes. Our findings indicate that merely increasing the size of encoders does not necessarily enhance the performance of VLMs. Moreover, we analyzed the effects of LLM backbone parameter size and data quality on the pretraining outcomes. Additionally, we explored the differences in scaling laws between LLMs and VLMs.

There has been a recent explosion of computer vision models which perform many tasks and are composed of an image encoder (usually a ViT) and an autoregressive decoder (usually a Transformer). However, most of this work simply presents one system and its results, leaving many questions regarding design decisions and trade-offs of such systems unanswered. In this work, we aim to provide such answers. We take a close look at autoregressive decoders for multi-task learning in multimodal computer vision, including classification, captioning, visual question answering, and optical character recognition. Through extensive systematic experiments, we study the effects of task and data mixture, training and regularization hyperparameters, conditioning type and specificity, modality combination, and more. Importantly, we compare these to well-tuned single-task baselines to highlight the cost incurred by multi-tasking. A key finding is that a small decoder learned on top of a frozen pretrained encoder works surprisingly well. We call this setup locked-image tuning with decoder (LiT-decoder). It can be seen as teaching a decoder to interact with a pretrained vision model via natural language.

Recent breakthroughs in vision-language models (VLMs) start a new page in the vision community. The VLMs provide stronger and more generalizable feature embeddings compared to those from ImageNet-pretrained models, thanks to the training on the large-scale Internet image-text pairs. However, despite the amazing achievement from the VLMs, vanilla Vision Transformers (ViTs) remain the default choice for the image encoder. Although pure transformer proves its effectiveness in the text encoding area, it remains questionable whether it is also the case for image encoding, especially considering that various types of networks are proposed on the ImageNet benchmark, which, unfortunately, are rarely studied in VLMs. Due to small data/model scale, the original conclusions of model design on ImageNet can be limited and biased. In this paper, we aim at building an evaluation protocol of vision models in the vision-language era under the contrastive language-image pretraining (CLIP) framework. We provide a comprehensive way to benchmark different vision models, covering their zero-shot performance and scalability in both model and training data sizes. To this end, we introduce ViTamin, a new vision models tailored for VLMs. ViTamin-L significantly outperforms ViT-L by 2.0% ImageNet zero-shot accuracy, when using the same publicly available DataComp-1B dataset and the same OpenCLIP training scheme. ViTamin-L presents promising results on 60 diverse benchmarks, including classification, retrieval, open-vocabulary detection and segmentation, and large multi-modal models. When further scaling up the model size, our ViTamin-XL with only 436M parameters attains 82.9% ImageNet zero-shot accuracy, surpassing 82.0% achieved by EVA-E that has ten times more parameters (4.4B).

We present in this paper a new architecture, named Convolutional vision Transformer (CvT), that improves Vision Transformer (ViT) in performance and efficiency by introducing convolutions into ViT to yield the best of both designs. This is accomplished through two primary modifications: a hierarchy of Transformers containing a new convolutional token embedding, and a convolutional Transformer block leveraging a convolutional projection. These changes introduce desirable properties of convolutional neural networks (CNNs) to the ViT architecture (\ie shift, scale, and distortion invariance) while maintaining the merits of Transformers (\ie dynamic attention, global context, and better generalization). We validate CvT by conducting extensive experiments, showing that this approach achieves state-of-the-art performance over other Vision Transformers and ResNets on ImageNet-1k, with fewer parameters and lower FLOPs. In addition, performance gains are maintained when pretrained on larger datasets (\eg ImageNet-22k) and fine-tuned to downstream tasks. Pre-trained on ImageNet-22k, our CvT-W24 obtains a top-1 accuracy of 87.7\% on the ImageNet-1k val set. Finally, our results show that the positional encoding, a crucial component in existing Vision Transformers, can be safely removed in our model, simplifying the design for higher resolution vision tasks. Code will be released at https://github.com/leoxiaobin/CvT.

Current large vision-language models (VLMs) often encounter challenges such as insufficient capabilities of a single visual component and excessively long visual tokens. These issues can limit the model's effectiveness in accurately interpreting complex visual information and over-lengthy contextual information. Addressing these challenges is crucial for enhancing the performance and applicability of VLMs. This paper proposes the use of ensemble experts technique to synergizes the capabilities of individual visual encoders, including those skilled in image-text matching, OCR, image segmentation, etc. This technique introduces a fusion network to unify the processing of outputs from different visual experts, while bridging the gap between image encoders and pre-trained LLMs. In addition, we explore different positional encoding schemes to alleviate the waste of positional encoding caused by lengthy image feature sequences, effectively addressing the issue of position overflow and length limitations. For instance, in our implementation, this technique significantly reduces the positional occupancy in models like SAM, from a substantial 4096 to a more efficient and manageable 64 or even down to 1. Experimental results demonstrate that VLMs with multiple experts exhibit consistently superior performance over isolated visual encoders and mark a significant performance boost as more experts are integrated. We have open-sourced the training code used in this report. All of these resources can be found on our project website.

We propose a conditional positional encoding (CPE) scheme for vision Transformers. Unlike previous fixed or learnable positional encodings, which are pre-defined and independent of input tokens, CPE is dynamically generated and conditioned on the local neighborhood of the input tokens. As a result, CPE can easily generalize to the input sequences that are longer than what the model has ever seen during training. Besides, CPE can keep the desired translation-invariance in the image classification task, resulting in improved performance. We implement CPE with a simple Position Encoding Generator (PEG) to get seamlessly incorporated into the current Transformer framework. Built on PEG, we present Conditional Position encoding Vision Transformer (CPVT). We demonstrate that CPVT has visually similar attention maps compared to those with learned positional encodings and delivers outperforming results. Our code is available at https://github.com/Meituan-AutoML/CPVT .

Training visual embeddings with labeled data supervision has been the de facto setup for representation learning in computer vision. Inspired by recent success of adopting masked image modeling (MIM) in self-supervised representation learning, we propose a simple yet effective setup that can easily integrate MIM into existing supervised training paradigms. In our design, in addition to the original classification task applied to a vision transformer image encoder, we add a shallow transformer-based decoder on top of the encoder and introduce an MIM task which tries to reconstruct image tokens based on masked image inputs. We show with minimal change in architecture and no overhead in inference that this setup is able to improve the quality of the learned representations for downstream tasks such as classification, image retrieval, and semantic segmentation. We conduct a comprehensive study and evaluation of our setup on public benchmarks. On ImageNet-1k, our ViT-B/14 model achieves 81.72% validation accuracy, 2.01% higher than the baseline model. On K-Nearest-Neighbor image retrieval evaluation with ImageNet-1k, the same model outperforms the baseline by 1.32%. We also show that this setup can be easily scaled to larger models and datasets. Code and checkpoints will be released.

Since the introduction of Vision Transformer (ViT), patchification has long been regarded as a de facto image tokenization approach for plain visual architectures. By compressing the spatial size of images, this approach can effectively shorten the token sequence and reduce the computational cost of ViT-like plain architectures. In this work, we aim to thoroughly examine the information loss caused by this patchification-based compressive encoding paradigm and how it affects visual understanding. We conduct extensive patch size scaling experiments and excitedly observe an intriguing scaling law in patchification: the models can consistently benefit from decreased patch sizes and attain improved predictive performance, until it reaches the minimum patch size of 1x1, i.e., pixel tokenization. This conclusion is broadly applicable across different vision tasks, various input scales, and diverse architectures such as ViT and the recent Mamba models. Moreover, as a by-product, we discover that with smaller patches, task-specific decoder heads become less critical for dense prediction. In the experiments, we successfully scale up the visual sequence to an exceptional length of 50,176 tokens, achieving a competitive test accuracy of 84.6% with a base-sized model on the ImageNet-1k benchmark. We hope this study can provide insights and theoretical foundations for future works of building non-compressive vision models. Code is available at https://github.com/wangf3014/Patch_Scaling.

In this paper, we present a novel methodology we call MDS-ViTNet (Multi Decoder Saliency by Vision Transformer Network) for enhancing visual saliency prediction or eye-tracking. This approach holds significant potential for diverse fields, including marketing, medicine, robotics, and retail. We propose a network architecture that leverages the Vision Transformer, moving beyond the conventional ImageNet backbone. The framework adopts an encoder-decoder structure, with the encoder utilizing a Swin transformer to efficiently embed most important features. This process involves a Transfer Learning method, wherein layers from the Vision Transformer are converted by the Encoder Transformer and seamlessly integrated into a CNN Decoder. This methodology ensures minimal information loss from the original input image. The decoder employs a multi-decoding technique, utilizing dual decoders to generate two distinct attention maps. These maps are subsequently combined into a singular output via an additional CNN model. Our trained model MDS-ViTNet achieves state-of-the-art results across several benchmarks. Committed to fostering further collaboration, we intend to make our code, models, and datasets accessible to the public.

Vision transformer based models bring significant improvements for image segmentation tasks. Although these architectures offer powerful capabilities irrespective of specific segmentation tasks, their use of computational resources can be taxing on deployed devices. One way to overcome this challenge is by adapting the computation level to the specific needs of the input image rather than the current one-size-fits-all approach. To this end, we introduce ECO-M2F or EffiCient TransfOrmer Encoders for Mask2Former-style models. Noting that the encoder module of M2F-style models incur high resource-intensive computations, ECO-M2F provides a strategy to self-select the number of hidden layers in the encoder, conditioned on the input image. To enable this self-selection ability for providing a balance between performance and computational efficiency, we present a three step recipe. The first step is to train the parent architecture to enable early exiting from the encoder. The second step is to create an derived dataset of the ideal number of encoder layers required for each training example. The third step is to use the aforementioned derived dataset to train a gating network that predicts the number of encoder layers to be used, conditioned on the input image. Additionally, to change the computational-accuracy tradeoff, only steps two and three need to be repeated which significantly reduces retraining time. Experiments on the public datasets show that the proposed approach reduces expected encoder computational cost while maintaining performance, adapts to various user compute resources, is flexible in architecture configurations, and can be extended beyond the segmentation task to object detection.

Very recently, a variety of vision transformer architectures for dense prediction tasks have been proposed and they show that the design of spatial attention is critical to their success in these tasks. In this work, we revisit the design of the spatial attention and demonstrate that a carefully-devised yet simple spatial attention mechanism performs favourably against the state-of-the-art schemes. As a result, we propose two vision transformer architectures, namely, Twins-PCPVT and Twins-SVT. Our proposed architectures are highly-efficient and easy to implement, only involving matrix multiplications that are highly optimized in modern deep learning frameworks. More importantly, the proposed architectures achieve excellent performance on a wide range of visual tasks, including image level classification as well as dense detection and segmentation. The simplicity and strong performance suggest that our proposed architectures may serve as stronger backbones for many vision tasks. Our code is released at https://github.com/Meituan-AutoML/Twins .

Vision-Language (VL) models have gained significant research focus, enabling remarkable advances in multimodal reasoning. These architectures typically comprise a vision encoder, a Large Language Model (LLM), and a projection module that aligns visual features with the LLM's representation space. Despite their success, a critical limitation persists: the vision encoding process remains decoupled from user queries, often in the form of image-related questions. Consequently, the resulting visual features may not be optimally attuned to the query-specific elements of the image. To address this, we introduce QA-ViT, a Question Aware Vision Transformer approach for multimodal reasoning, which embeds question awareness directly within the vision encoder. This integration results in dynamic visual features focusing on relevant image aspects to the posed question. QA-ViT is model-agnostic and can be incorporated efficiently into any VL architecture. Extensive experiments demonstrate the effectiveness of applying our method to various multimodal architectures, leading to consistent improvement across diverse tasks and showcasing its potential for enhancing visual and scene-text understanding.

The Vision Transformer architecture has shown to be competitive in the computer vision (CV) space where it has dethroned convolution-based networks in several benchmarks. Nevertheless, convolutional neural networks (CNN) remain the preferential architecture for the representation module in reinforcement learning. In this work, we study pretraining a Vision Transformer using several state-of-the-art self-supervised methods and assess the quality of the learned representations. To show the importance of the temporal dimension in this context we propose an extension of VICReg to better capture temporal relations between observations by adding a temporal order verification task. Our results show that all methods are effective in learning useful representations and avoiding representational collapse for observations from Atari Learning Environment (ALE) which leads to improvements in data efficiency when we evaluated in reinforcement learning (RL). Moreover, the encoder pretrained with the temporal order verification task shows the best results across all experiments, with richer representations, more focused attention maps and sparser representation vectors throughout the layers of the encoder, which shows the importance of exploring such similarity dimension. With this work, we hope to provide some insights into the representations learned by ViT during a self-supervised pretraining with observations from RL environments and which properties arise in the representations that lead to the best-performing agents. The source code will be available at: https://github.com/mgoulao/TOV-VICReg

Vision Transformers (ViTs) have emerged as the dominant architecture for visual processing tasks, demonstrating excellent scalability with increased training data and model size. However, recent work has identified the emergence of artifact tokens in ViTs that are incongruous with the local semantics. These anomalous tokens degrade ViT performance in tasks that require fine-grained localization or structural coherence. An effective mitigation of this issue is to the addition of register tokens to ViTs, which implicitly "absorb" the artifact term during training. Given the availability of various large-scale pre-trained ViTs, in this paper we aim at equipping them with such register tokens without the need of re-training them from scratch, which is infeasible considering their size. Specifically, we propose Post Hoc Registers (PH-Reg), an efficient self-distillation method that integrates registers into an existing ViT without requiring additional labeled data and full retraining. PH-Reg initializes both teacher and student networks from the same pre-trained ViT. The teacher remains frozen and unmodified, while the student is augmented with randomly initialized register tokens. By applying test-time augmentation to the teacher's inputs, we generate denoised dense embeddings free of artifacts, which are then used to optimize only a small subset of unlocked student weights. We show that our approach can effectively reduce the number of artifact tokens, improving the segmentation and depth prediction of the student ViT under zero-shot and linear probing.

Recent advances in scene understanding benefit a lot from depth maps because of the 3D geometry information, especially in complex conditions (e.g., low light and overexposed). Existing approaches encode depth maps along with RGB images and perform feature fusion between them to enable more robust predictions. Taking into account that depth can be regarded as a geometry supplement for RGB images, a straightforward question arises: Do we really need to explicitly encode depth information with neural networks as done for RGB images? Based on this insight, in this paper, we investigate a new way to learn RGBD feature representations and present DFormerv2, a strong RGBD encoder that explicitly uses depth maps as geometry priors rather than encoding depth information with neural networks. Our goal is to extract the geometry clues from the depth and spatial distances among all the image patch tokens, which will then be used as geometry priors to allocate attention weights in self-attention. Extensive experiments demonstrate that DFormerv2 exhibits exceptional performance in various RGBD semantic segmentation benchmarks. Code is available at: https://github.com/VCIP-RGBD/DFormer.

Vision Transformer (ViT) has emerged as a powerful architecture in the realm of modern computer vision. However, its application in certain imaging fields, such as microscopy and satellite imaging, presents unique challenges. In these domains, images often contain multiple channels, each carrying semantically distinct and independent information. Furthermore, the model must demonstrate robustness to sparsity in input channels, as they may not be densely available during training or testing. In this paper, we propose a modification to the ViT architecture that enhances reasoning across the input channels and introduce Hierarchical Channel Sampling (HCS) as an additional regularization technique to ensure robustness when only partial channels are presented during test time. Our proposed model, ChannelViT, constructs patch tokens independently from each input channel and utilizes a learnable channel embedding that is added to the patch tokens, similar to positional embeddings. We evaluate the performance of ChannelViT on ImageNet, JUMP-CP (microscopy cell imaging), and So2Sat (satellite imaging). Our results show that ChannelViT outperforms ViT on classification tasks and generalizes well, even when a subset of input channels is used during testing. Across our experiments, HCS proves to be a powerful regularizer, independent of the architecture employed, suggesting itself as a straightforward technique for robust ViT training. Lastly, we find that ChannelViT generalizes effectively even when there is limited access to all channels during training, highlighting its potential for multi-channel imaging under real-world conditions with sparse sensors. Our code is available at https://github.com/insitro/ChannelViT.

The Vision Transformer (ViT) has made significant strides in the field of computer vision. However, as the depth of the model and the resolution of the input images increase, the computational cost associated with training and running ViT models has surged dramatically. This paper proposes a hybrid model based on CNN and Vision Transformer, named CI2P-ViT. The model incorporates a module called CI2P, which utilizes the CompressAI encoder to compress images and subsequently generates a sequence of patches through a series of convolutions. CI2P can replace the Patch Embedding component in the ViT model, enabling seamless integration into existing ViT models. Compared to ViT-B/16, CI2P-ViT has the number of patches input to the self-attention layer reduced to a quarter of the original. This design not only significantly reduces the computational cost of the ViT model but also effectively enhances the model's accuracy by introducing the inductive bias properties of CNN. The ViT model's precision is markedly enhanced. When trained from the ground up on the Animals-10 dataset, CI2P-ViT achieved an accuracy rate of 92.37%, representing a 3.3% improvement over the ViT-B/16 baseline. Additionally, the model's computational operations, measured in floating-point operations per second (FLOPs), were diminished by 63.35%, and it exhibited a 2-fold increase in training velocity on identical hardware configurations.

Multimodal Large Language Models have advanced AI in applications like text-to-video generation and visual question answering. These models rely on visual encoders to convert non-text data into vectors, but current encoders either lack semantic alignment or overlook non-salient objects. We propose the Guiding Visual Encoder to Perceive Overlooked Information (GiVE) approach. GiVE enhances visual representation with an Attention-Guided Adapter (AG-Adapter) module and an Object-focused Visual Semantic Learning module. These incorporate three novel loss terms: Object-focused Image-Text Contrast (OITC) loss, Object-focused Image-Image Contrast (OIIC) loss, and Object-focused Image Discrimination (OID) loss, improving object consideration, retrieval accuracy, and comprehensiveness. Our contributions include dynamic visual focus adjustment, novel loss functions to enhance object retrieval, and the Multi-Object Instruction (MOInst) dataset. Experiments show our approach achieves state-of-the-art performance.

Building on the advances of language models, Large Multimodal Models (LMMs) have contributed significant improvements in video understanding. While the current video LMMs utilize advanced Large Language Models (LLMs), they rely on either image or video encoders to process visual inputs, each of which has its own limitations. Image encoders excel at capturing rich spatial details from frame sequences but lack explicit temporal context, which can be important in videos with intricate action sequences. On the other hand, video encoders provide temporal context but are often limited by computational constraints that lead to processing only sparse frames at lower resolutions, resulting in reduced contextual and spatial understanding. To this end, we introduce VideoGPT+, which combines the complementary benefits of the image encoder (for detailed spatial understanding) and the video encoder (for global temporal context modeling). The model processes videos by dividing them into smaller segments and applies an adaptive pooling strategy on features extracted by both image and video encoders. Our architecture showcases improved performance across multiple video benchmarks, including VCGBench, MVBench and Zero-shot question-answering. Further, we develop 112K video-instruction set using a novel semi-automatic annotation pipeline which further improves the model performance. Additionally, to comprehensively evaluate video LMMs, we present VCGBench-Diverse, covering 18 broad video categories such as lifestyle, sports, science, gaming, and surveillance videos. This benchmark with 4,354 question-answer pairs evaluates the generalization of existing LMMs on dense video captioning, spatial and temporal understanding, and complex reasoning, ensuring comprehensive assessment across diverse video types and dynamics. Code: https://github.com/mbzuai-oryx/VideoGPT-plus.

We propose the Large View Synthesis Model (LVSM), a novel transformer-based approach for scalable and generalizable novel view synthesis from sparse-view inputs. We introduce two architectures: (1) an encoder-decoder LVSM, which encodes input image tokens into a fixed number of 1D latent tokens, functioning as a fully learned scene representation, and decodes novel-view images from them; and (2) a decoder-only LVSM, which directly maps input images to novel-view outputs, completely eliminating intermediate scene representations. Both models bypass the 3D inductive biases used in previous methods -- from 3D representations (e.g., NeRF, 3DGS) to network designs (e.g., epipolar projections, plane sweeps) -- addressing novel view synthesis with a fully data-driven approach. While the encoder-decoder model offers faster inference due to its independent latent representation, the decoder-only LVSM achieves superior quality, scalability, and zero-shot generalization, outperforming previous state-of-the-art methods by 1.5 to 3.5 dB PSNR. Comprehensive evaluations across multiple datasets demonstrate that both LVSM variants achieve state-of-the-art novel view synthesis quality. Notably, our models surpass all previous methods even with reduced computational resources (1-2 GPUs). Please see our website for more details: https://haian-jin.github.io/projects/LVSM/ .

Visual search is important in our daily life. The efficient allocation of visual attention is critical to effectively complete visual search tasks. Prior research has predominantly modelled the spatial allocation of visual attention in images at the pixel level, e.g. using a saliency map. However, emerging evidence shows that visual attention is guided by objects rather than pixel intensities. This paper introduces the Object-level Attention Transformer (OAT), which predicts human scanpaths as they search for a target object within a cluttered scene of distractors. OAT uses an encoder-decoder architecture. The encoder captures information about the position and appearance of the objects within an image and about the target. The decoder predicts the gaze scanpath as a sequence of object fixations, by integrating output features from both the encoder and decoder. We also propose a new positional encoding that better reflects spatial relationships between objects. We evaluated OAT on the Amazon book cover dataset and a new dataset for visual search that we collected. OAT's predicted gaze scanpaths align more closely with human gaze patterns, compared to predictions by algorithms based on spatial attention on both established metrics and a novel behavioural-based metric. Our results demonstrate the generalization ability of OAT, as it accurately predicts human scanpaths for unseen layouts and target objects.

This paper investigates the capability of plain Vision Transformers (ViTs) for semantic segmentation using the encoder-decoder framework and introduces SegViTv2. In this study, we introduce a novel Attention-to-Mask (\atm) module to design a lightweight decoder effective for plain ViT. The proposed ATM converts the global attention map into semantic masks for high-quality segmentation results. Our decoder outperforms the popular decoder UPerNet using various ViT backbones while consuming only about 5% of the computational cost. For the encoder, we address the concern of the relatively high computational cost in the ViT-based encoders and propose a Shrunk++ structure that incorporates edge-aware query-based down-sampling (EQD) and query-based upsampling (QU) modules. The Shrunk++ structure reduces the computational cost of the encoder by up to 50% while maintaining competitive performance. Furthermore, we propose to adapt SegViT for continual semantic segmentation, demonstrating nearly zero forgetting of previously learned knowledge. Experiments show that our proposed SegViTv2 surpasses recent segmentation methods on three popular benchmarks including ADE20k, COCO-Stuff-10k and PASCAL-Context datasets. The code is available through the following link: https://github.com/zbwxp/SegVit.

Neural fields, also known as implicit neural representations (INRs), have shown a remarkable capability of representing, generating, and manipulating various data types, allowing for continuous data reconstruction at a low memory footprint. Though promising, INRs applied to video compression still need to improve their rate-distortion performance by a large margin, and require a huge number of parameters and long training iterations to capture high-frequency details, limiting their wider applicability. Resolving this problem remains a quite challenging task, which would make INRs more accessible in compression tasks. We take a step towards resolving these shortcomings by introducing neural representations for videos NeRV++, an enhanced implicit neural video representation, as more straightforward yet effective enhancement over the original NeRV decoder architecture, featuring separable conv2d residual blocks (SCRBs) that sandwiches the upsampling block (UB), and a bilinear interpolation skip layer for improved feature representation. NeRV++ allows videos to be directly represented as a function approximated by a neural network, and significantly enhance the representation capacity beyond current INR-based video codecs. We evaluate our method on UVG, MCL JVC, and Bunny datasets, achieving competitive results for video compression with INRs. This achievement narrows the gap to autoencoder-based video coding, marking a significant stride in INR-based video compression research.

In this paper, we present a light-weight detection transformer, LW-DETR, which outperforms YOLOs for real-time object detection. The architecture is a simple stack of a ViT encoder, a projector, and a shallow DETR decoder. Our approach leverages recent advanced techniques, such as training-effective techniques, e.g., improved loss and pretraining, and interleaved window and global attentions for reducing the ViT encoder complexity. We improve the ViT encoder by aggregating multi-level feature maps, and the intermediate and final feature maps in the ViT encoder, forming richer feature maps, and introduce window-major feature map organization for improving the efficiency of interleaved attention computation. Experimental results demonstrate that the proposed approach is superior over existing real-time detectors, e.g., YOLO and its variants, on COCO and other benchmark datasets. Code and models are available at (https://github.com/Atten4Vis/LW-DETR).

Vision Transformers (ViTs) have attracted a lot of popularity in recent years, due to their exceptional capabilities in modeling long-range spatial dependencies and scalability for large scale training. Although the training parallelism of self-attention mechanism plays an important role in retaining great performance, its quadratic complexity baffles the application of ViTs in many scenarios which demand fast inference. This effect is even more pronounced in applications in which autoregressive modeling of input features is required. In Natural Language Processing (NLP), a new stream of efforts have proposed parallelizable models with recurrent formulation that allows for efficient inference in generative applications. Inspired by this trend, we propose a new class of computer vision models, dubbed Vision Retention Networks (ViR), with dual parallel and recurrent formulations, which strike an optimal balance between fast inference and parallel training with competitive performance. In particular, ViR scales favorably for image throughput and memory consumption in tasks that require higher-resolution images due to its flexible formulation in processing large sequence lengths. The ViR is the first attempt to realize dual parallel and recurrent equivalency in a general vision backbone for recognition tasks. We have validated the effectiveness of ViR through extensive experiments with different dataset sizes and various image resolutions and achieved competitive performance. Our code and pretrained models will be made publicly available.

Vision-language models have made significant strides recently, demonstrating superior performance across a range of tasks, e.g. optical character recognition and complex diagram analysis. Building on this trend, we introduce a new vision-language model, POINTS1.5, designed to excel in various real-world applications. POINTS1.5 is an enhancement of POINTS1.0 and incorporates several key innovations: i) We replace the original CLIP vision encoder, which had a fixed image resolution, with a NaViT-style vision encoder that supports native dynamic high resolution. This allows POINTS1.5 to process images of any resolution without needing to split them into tiles. ii) We add bilingual support to POINTS1.5, significantly enhancing its capability in Chinese. Due to the scarcity of open-source Chinese datasets for vision-language models, we collect numerous images from the Internet and annotate them using a combination of manual and automatic methods. iii) We propose a set of rigorous filtering methods for visual instruction tuning datasets. We comprehensively evaluate all these filtering methods, and choose the most effective ones to obtain the final visual instruction tuning set. Thanks to these innovations, POINTS1.5 significantly outperforms POINTS1.0 and demonstrates strong performance across a range of real-world applications. Notably, POINTS1.5-7B is trained on fewer than 4 billion tokens and ranks first on the OpenCompass leaderboard among models with fewer than 10 billion parameters

A central objective in computer vision is to design models with appropriate 2-D inductive bias. Desiderata for 2D inductive bias include two-dimensional position awareness, dynamic spatial locality, and translation and permutation invariance. To address these goals, we leverage an expressive variation of the multidimensional State Space Model (SSM). Our approach introduces efficient parameterization, accelerated computation, and a suitable normalization scheme. Empirically, we observe that incorporating our layer at the beginning of each transformer block of Vision Transformers (ViT) significantly enhances performance for multiple ViT backbones and across datasets. The new layer is effective even with a negligible amount of additional parameters and inference time. Ablation studies and visualizations demonstrate that the layer has a strong 2-D inductive bias. For example, vision transformers equipped with our layer exhibit effective performance even without positional encoding

Vision Transformers (ViT) have marked a paradigm shift in computer vision, outperforming state-of-the-art models across diverse tasks. However, their practical deployment is hampered by high computational and memory demands. This study addresses the challenge by evaluating four primary model compression techniques: quantization, low-rank approximation, knowledge distillation, and pruning. We methodically analyze and compare the efficacy of these techniques and their combinations in optimizing ViTs for resource-constrained environments. Our comprehensive experimental evaluation demonstrates that these methods facilitate a balanced compromise between model accuracy and computational efficiency, paving the way for wider application in edge computing devices.

We present a novel masked image modeling (MIM) approach, context autoencoder (CAE), for self-supervised representation pretraining. We pretrain an encoder by making predictions in the encoded representation space. The pretraining tasks include two tasks: masked representation prediction - predict the representations for the masked patches, and masked patch reconstruction - reconstruct the masked patches. The network is an encoder-regressor-decoder architecture: the encoder takes the visible patches as input; the regressor predicts the representations of the masked patches, which are expected to be aligned with the representations computed from the encoder, using the representations of visible patches and the positions of visible and masked patches; the decoder reconstructs the masked patches from the predicted encoded representations. The CAE design encourages the separation of learning the encoder (representation) from completing the pertaining tasks: masked representation prediction and masked patch reconstruction tasks, and making predictions in the encoded representation space empirically shows the benefit to representation learning. We demonstrate the effectiveness of our CAE through superior transfer performance in downstream tasks: semantic segmentation, object detection and instance segmentation, and classification. The code will be available at https://github.com/Atten4Vis/CAE.

Vision Transformers (ViTs) have shown remarkable performance and scalability across various computer vision tasks. To apply single-scale ViTs to image segmentation, existing methods adopt a convolutional adapter to generate multi-scale features, a pixel decoder to fuse these features, and a Transformer decoder that uses the fused features to make predictions. In this paper, we show that the inductive biases introduced by these task-specific components can instead be learned by the ViT itself, given sufficiently large models and extensive pre-training. Based on these findings, we introduce the Encoder-only Mask Transformer (EoMT), which repurposes the plain ViT architecture to conduct image segmentation. With large-scale models and pre-training, EoMT obtains a segmentation accuracy similar to state-of-the-art models that use task-specific components. At the same time, EoMT is significantly faster than these methods due to its architectural simplicity, e.g., up to 4x faster with ViT-L. Across a range of model sizes, EoMT demonstrates an optimal balance between segmentation accuracy and prediction speed, suggesting that compute resources are better spent on scaling the ViT itself rather than adding architectural complexity. Code: https://www.tue-mps.org/eomt/.

Convolutional networks for single-view object reconstruction have shown impressive performance and have become a popular subject of research. All existing techniques are united by the idea of having an encoder-decoder network that performs non-trivial reasoning about the 3D structure of the output space. In this work, we set up two alternative approaches that perform image classification and retrieval respectively. These simple baselines yield better results than state-of-the-art methods, both qualitatively and quantitatively. We show that encoder-decoder methods are statistically indistinguishable from these baselines, thus indicating that the current state of the art in single-view object reconstruction does not actually perform reconstruction but image classification. We identify aspects of popular experimental procedures that elicit this behavior and discuss ways to improve the current state of research.

Chart understanding requires models to effectively analyze and reason about numerical data, textual elements, and complex visual components. Our observations reveal that the perception capabilities of existing large vision-language models (LVLMs) constitute a critical bottleneck in this process. In this study, we delve into this perception bottleneck by decomposing it into two components: the vision encoder bottleneck, where the visual representation may fail to encapsulate the correct information, and the extraction bottleneck, where the language model struggles to extract the necessary information from the provided visual representations. Through comprehensive experiments, we find that (1) the information embedded within visual representations is substantially richer than what is typically captured by linear extractors, such as the widely used retrieval accuracy metric; (2) While instruction tuning effectively enhances the extraction capability of LVLMs, the vision encoder remains a critical bottleneck, demanding focused attention and improvement. Therefore, we further enhance the visual encoder to mitigate the vision encoder bottleneck under a contrastive learning framework. Empirical results demonstrate that our approach significantly mitigates the perception bottleneck and improves the ability of LVLMs to comprehend charts. Code is publicly available at https://github.com/hkust-nlp/Vision4Chart.

A Vision Transformer (ViT) is a simple neural architecture amenable to serve several computer vision tasks. It has limited built-in architectural priors, in contrast to more recent architectures that incorporate priors either about the input data or of specific tasks. Recent works show that ViTs benefit from self-supervised pre-training, in particular BerT-like pre-training like BeiT. In this paper, we revisit the supervised training of ViTs. Our procedure builds upon and simplifies a recipe introduced for training ResNet-50. It includes a new simple data-augmentation procedure with only 3 augmentations, closer to the practice in self-supervised learning. Our evaluations on Image classification (ImageNet-1k with and without pre-training on ImageNet-21k), transfer learning and semantic segmentation show that our procedure outperforms by a large margin previous fully supervised training recipes for ViT. It also reveals that the performance of our ViT trained with supervision is comparable to that of more recent architectures. Our results could serve as better baselines for recent self-supervised approaches demonstrated on ViT.

Vision Transformers have achieved great success in computer visions, delivering exceptional performance across various tasks. However, their inherent reliance on sequential input enforces the manual partitioning of images into patch sequences, which disrupts the image's inherent structural and semantic continuity. To handle this, we propose a novel Pattern Transformer (Patternformer) to adaptively convert images to pattern sequences for Transformer input. Specifically, we employ the Convolutional Neural Network to extract various patterns from the input image, with each channel representing a unique pattern that is fed into the succeeding Transformer as a visual token. By enabling the network to optimize these patterns, each pattern concentrates on its local region of interest, thereby preserving its intrinsic structural and semantic information. Only employing the vanilla ResNet and Transformer, we have accomplished state-of-the-art performance on CIFAR-10 and CIFAR-100, and have achieved competitive results on ImageNet.

A large-scale labeled dataset is a key factor for the success of supervised deep learning in computer vision. However, a limited number of annotated data is very common, especially in ophthalmic image analysis, since manual annotation is time-consuming and labor-intensive. Self-supervised learning (SSL) methods bring huge opportunities for better utilizing unlabeled data, as they do not need massive annotations. With an attempt to use as many as possible unlabeled ophthalmic images, it is necessary to break the dimension barrier, simultaneously making use of both 2D and 3D images. In this paper, we propose a universal self-supervised Transformer framework, named Uni4Eye, to discover the inherent image property and capture domain-specific feature embedding in ophthalmic images. Uni4Eye can serve as a global feature extractor, which builds its basis on a Masked Image Modeling task with a Vision Transformer (ViT) architecture. We employ a Unified Patch Embedding module to replace the origin patch embedding module in ViT for jointly processing both 2D and 3D input images. Besides, we design a dual-branch multitask decoder module to simultaneously perform two reconstruction tasks on the input image and its gradient map, delivering discriminative representations for better convergence. We evaluate the performance of our pre-trained Uni4Eye encoder by fine-tuning it on six downstream ophthalmic image classification tasks. The superiority of Uni4Eye is successfully established through comparisons to other state-of-the-art SSL pre-training methods.

In this paper, we address the challenges posed by the substantial training time and memory consumption associated with video transformers, focusing on the ViViT (Video Vision Transformer) model, in particular the Factorised Encoder version, as our baseline for action recognition tasks. The factorised encoder variant follows the late-fusion approach that is adopted by many state of the art approaches. Despite standing out for its favorable speed/accuracy tradeoffs among the different variants of ViViT, its considerable training time and memory requirements still pose a significant barrier to entry. Our method is designed to lower this barrier and is based on the idea of freezing the spatial transformer during training. This leads to a low accuracy model if naively done. But we show that by (1) appropriately initializing the temporal transformer (a module responsible for processing temporal information) (2) introducing a compact adapter model connecting frozen spatial representations ((a module that selectively focuses on regions of the input image) to the temporal transformer, we can enjoy the benefits of freezing the spatial transformer without sacrificing accuracy. Through extensive experimentation over 6 benchmarks, we demonstrate that our proposed training strategy significantly reduces training costs (by sim 50%) and memory consumption while maintaining or slightly improving performance by up to 1.79\% compared to the baseline model. Our approach additionally unlocks the capability to utilize larger image transformer models as our spatial transformer and access more frames with the same memory consumption.

Encoder-free architectures have been preliminarily explored in the 2D visual domain, yet it remains an open question whether they can be effectively applied to 3D understanding scenarios. In this paper, we present the first comprehensive investigation into the potential of encoder-free architectures to overcome the challenges of encoder-based 3D Large Multimodal Models (LMMs). These challenges include the failure to adapt to varying point cloud resolutions and the point features from the encoder not meeting the semantic needs of Large Language Models (LLMs). We identify key aspects for 3D LMMs to remove the encoder and enable the LLM to assume the role of the 3D encoder: 1) We propose the LLM-embedded Semantic Encoding strategy in the pre-training stage, exploring the effects of various point cloud self-supervised losses. And we present the Hybrid Semantic Loss to extract high-level semantics. 2) We introduce the Hierarchical Geometry Aggregation strategy in the instruction tuning stage. This incorporates inductive bias into the LLM early layers to focus on the local details of the point clouds. To the end, we present the first Encoder-free 3D LMM, ENEL. Our 7B model rivals the current state-of-the-art model, ShapeLLM-13B, achieving 55.0%, 50.92%, and 42.7% on the classification, captioning, and VQA tasks, respectively. Our results demonstrate that the encoder-free architecture is highly promising for replacing encoder-based architectures in the field of 3D understanding. The code is released at https://github.com/Ivan-Tang-3D/ENEL

Transformers have recently emerged as a powerful tool for learning visual representations. In this paper, we identify and characterize artifacts in feature maps of both supervised and self-supervised ViT networks. The artifacts correspond to high-norm tokens appearing during inference primarily in low-informative background areas of images, that are repurposed for internal computations. We propose a simple yet effective solution based on providing additional tokens to the input sequence of the Vision Transformer to fill that role. We show that this solution fixes that problem entirely for both supervised and self-supervised models, sets a new state of the art for self-supervised visual models on dense visual prediction tasks, enables object discovery methods with larger models, and most importantly leads to smoother feature maps and attention maps for downstream visual processing.

Vision Transformers (ViT) have been shown to attain highly competitive performance for a wide range of vision applications, such as image classification, object detection and semantic image segmentation. In comparison to convolutional neural networks, the Vision Transformer's weaker inductive bias is generally found to cause an increased reliance on model regularization or data augmentation ("AugReg" for short) when training on smaller training datasets. We conduct a systematic empirical study in order to better understand the interplay between the amount of training data, AugReg, model size and compute budget. As one result of this study we find that the combination of increased compute and AugReg can yield models with the same performance as models trained on an order of magnitude more training data: we train ViT models of various sizes on the public ImageNet-21k dataset which either match or outperform their counterparts trained on the larger, but not publicly available JFT-300M dataset.

A significant weakness of most current deep Convolutional Neural Networks is the need to train them using vast amounts of manu- ally labelled data. In this work we propose a unsupervised framework to learn a deep convolutional neural network for single view depth predic- tion, without requiring a pre-training stage or annotated ground truth depths. We achieve this by training the network in a manner analogous to an autoencoder. At training time we consider a pair of images, source and target, with small, known camera motion between the two such as a stereo pair. We train the convolutional encoder for the task of predicting the depth map for the source image. To do so, we explicitly generate an inverse warp of the target image using the predicted depth and known inter-view displacement, to reconstruct the source image; the photomet- ric error in the reconstruction is the reconstruction loss for the encoder. The acquisition of this training data is considerably simpler than for equivalent systems, requiring no manual annotation, nor calibration of depth sensor to camera. We show that our network trained on less than half of the KITTI dataset (without any further augmentation) gives com- parable performance to that of the state of art supervised methods for single view depth estimation.

Although convolutional networks have been the dominant architecture for vision tasks for many years, recent experiments have shown that Transformer-based models, most notably the Vision Transformer (ViT), may exceed their performance in some settings. However, due to the quadratic runtime of the self-attention layers in Transformers, ViTs require the use of patch embeddings, which group together small regions of the image into single input features, in order to be applied to larger image sizes. This raises a question: Is the performance of ViTs due to the inherently-more-powerful Transformer architecture, or is it at least partly due to using patches as the input representation? In this paper, we present some evidence for the latter: specifically, we propose the ConvMixer, an extremely simple model that is similar in spirit to the ViT and the even-more-basic MLP-Mixer in that it operates directly on patches as input, separates the mixing of spatial and channel dimensions, and maintains equal size and resolution throughout the network. In contrast, however, the ConvMixer uses only standard convolutions to achieve the mixing steps. Despite its simplicity, we show that the ConvMixer outperforms the ViT, MLP-Mixer, and some of their variants for similar parameter counts and data set sizes, in addition to outperforming classical vision models such as the ResNet. Our code is available at https://github.com/locuslab/convmixer.

We introduce an unsupervised feature learning approach that embeds 3D shape information into a single-view image representation. The main idea is a self-supervised training objective that, given only a single 2D image, requires all unseen views of the object to be predictable from learned features. We implement this idea as an encoder-decoder convolutional neural network. The network maps an input image of an unknown category and unknown viewpoint to a latent space, from which a deconvolutional decoder can best "lift" the image to its complete viewgrid showing the object from all viewing angles. Our class-agnostic training procedure encourages the representation to capture fundamental shape primitives and semantic regularities in a data-driven manner---without manual semantic labels. Our results on two widely-used shape datasets show 1) our approach successfully learns to perform "mental rotation" even for objects unseen during training, and 2) the learned latent space is a powerful representation for object recognition, outperforming several existing unsupervised feature learning methods.

Implicit neural representations store videos as neural networks and have performed well for various vision tasks such as video compression and denoising. With frame index or positional index as input, implicit representations (NeRV, E-NeRV, \etc) reconstruct video from fixed and content-agnostic embeddings. Such embedding largely limits the regression capacity and internal generalization for video interpolation. In this paper, we propose a Hybrid Neural Representation for Videos (HNeRV), where a learnable encoder generates content-adaptive embeddings, which act as the decoder input. Besides the input embedding, we introduce HNeRV blocks, which ensure model parameters are evenly distributed across the entire network, such that higher layers (layers near the output) can have more capacity to store high-resolution content and video details. With content-adaptive embeddings and re-designed architecture, HNeRV outperforms implicit methods in video regression tasks for both reconstruction quality (+4.7 PSNR) and convergence speed (16times faster), and shows better internal generalization. As a simple and efficient video representation, HNeRV also shows decoding advantages for speed, flexibility, and deployment, compared to traditional codecs~(H.264, H.265) and learning-based compression methods. Finally, we explore the effectiveness of HNeRV on downstream tasks such as video compression and video inpainting. We provide project page at https://haochen-rye.github.io/HNeRV, and Code at https://github.com/haochen-rye/HNeRV

Customization of text-to-image models enables users to insert custom concepts and generate the concepts in unseen settings. Existing methods either rely on costly test-time optimization or train encoders on single-image training datasets without multi-image supervision, leading to worse image quality. We propose a simple approach that addresses both limitations. We first leverage existing text-to-image models and 3D datasets to create a high-quality Synthetic Customization Dataset (SynCD) consisting of multiple images of the same object in different lighting, backgrounds, and poses. We then propose a new encoder architecture based on shared attention mechanisms that better incorporate fine-grained visual details from input images. Finally, we propose a new inference technique that mitigates overexposure issues during inference by normalizing the text and image guidance vectors. Through extensive experiments, we show that our model, trained on the synthetic dataset with the proposed encoder and inference algorithm, outperforms existing tuning-free methods on standard customization benchmarks.

his paper tackles a significant challenge faced by Vision Transformers (ViTs): their constrained scalability across different image resolutions. Typically, ViTs experience a performance decline when processing resolutions different from those seen during training. Our work introduces two key innovations to address this issue. Firstly, we propose a novel module for dynamic resolution adjustment, designed with a single Transformer block, specifically to achieve highly efficient incremental token integration. Secondly, we introduce fuzzy positional encoding in the Vision Transformer to provide consistent positional awareness across multiple resolutions, thereby preventing overfitting to any single training resolution. Our resulting model, ViTAR (Vision Transformer with Any Resolution), demonstrates impressive adaptability, achieving 83.3\% top-1 accuracy at a 1120x1120 resolution and 80.4\% accuracy at a 4032x4032 resolution, all while reducing computational costs. ViTAR also shows strong performance in downstream tasks such as instance and semantic segmentation and can easily combined with self-supervised learning techniques like Masked AutoEncoder. Our work provides a cost-effective solution for enhancing the resolution scalability of ViTs, paving the way for more versatile and efficient high-resolution image processing.

Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs) stand as the two most popular foundation models for visual representation learning. While CNNs exhibit remarkable scalability with linear complexity w.r.t. image resolution, ViTs surpass them in fitting capabilities despite contending with quadratic complexity. A closer inspection reveals that ViTs achieve superior visual modeling performance through the incorporation of global receptive fields and dynamic weights. This observation motivates us to propose a novel architecture that inherits these components while enhancing computational efficiency. To this end, we draw inspiration from the recently introduced state space model and propose the Visual State Space Model (VMamba), which achieves linear complexity without sacrificing global receptive fields. To address the encountered direction-sensitive issue, we introduce the Cross-Scan Module (CSM) to traverse the spatial domain and convert any non-causal visual image into order patch sequences. Extensive experimental results substantiate that VMamba not only demonstrates promising capabilities across various visual perception tasks, but also exhibits more pronounced advantages over established benchmarks as the image resolution increases. Source code has been available at https://github.com/MzeroMiko/VMamba.

Shift equivariance is a fundamental principle that governs how we perceive the world - our recognition of an object remains invariant with respect to shifts. Transformers have gained immense popularity due to their effectiveness in both language and vision tasks. While the self-attention operator in vision transformers (ViT) is permutation-equivariant and thus shift-equivariant, patch embedding, positional encoding, and subsampled attention in ViT variants can disrupt this property, resulting in inconsistent predictions even under small shift perturbations. Although there is a growing trend in incorporating the inductive bias of convolutional neural networks (CNNs) into vision transformers, it does not fully address the issue. We propose an adaptive polyphase anchoring algorithm that can be seamlessly integrated into vision transformer models to ensure shift-equivariance in patch embedding and subsampled attention modules, such as window attention and global subsampled attention. Furthermore, we utilize depth-wise convolution to encode positional information. Our algorithms enable ViT, and its variants such as Twins to achieve 100% consistency with respect to input shift, demonstrate robustness to cropping, flipping, and affine transformations, and maintain consistent predictions even when the original models lose 20 percentage points on average when shifted by just a few pixels with Twins' accuracy dropping from 80.57% to 62.40%.

Vision Transformers (ViT) have emerged as the de-facto choice for numerous industry grade vision solutions. But their inference cost can be prohibitive for many settings, as they compute self-attention in each layer which suffers from quadratic computational complexity in the number of tokens. On the other hand, spatial information in images and spatio-temporal information in videos is usually sparse and redundant. In this work, we introduce LookupViT, that aims to exploit this information sparsity to reduce ViT inference cost. LookupViT provides a novel general purpose vision transformer block that operates by compressing information from higher resolution tokens to a fixed number of tokens. These few compressed tokens undergo meticulous processing, while the higher-resolution tokens are passed through computationally cheaper layers. Information sharing between these two token sets is enabled through a bidirectional cross-attention mechanism. The approach offers multiple advantages - (a) easy to implement on standard ML accelerators (GPUs/TPUs) via standard high-level operators, (b) applicable to standard ViT and its variants, thus generalizes to various tasks, (c) can handle different tokenization and attention approaches. LookupViT also offers flexibility for the compressed tokens, enabling performance-computation trade-offs in a single trained model. We show LookupViT's effectiveness on multiple domains - (a) for image-classification (ImageNet-1K and ImageNet-21K), (b) video classification (Kinetics400 and Something-Something V2), (c) image captioning (COCO-Captions) with a frozen encoder. LookupViT provides 2times reduction in FLOPs while upholding or improving accuracy across these domains. In addition, LookupViT also demonstrates out-of-the-box robustness and generalization on image classification (ImageNet-C,R,A,O), improving by up to 4% over ViT.

Vision Transformer (ViT) extends the application range of transformers from language processing to computer vision tasks as being an alternative architecture against the existing convolutional neural networks (CNN). Since the transformer-based architecture has been innovative for computer vision modeling, the design convention towards an effective architecture has been less studied yet. From the successful design principles of CNN, we investigate the role of spatial dimension conversion and its effectiveness on transformer-based architecture. We particularly attend to the dimension reduction principle of CNNs; as the depth increases, a conventional CNN increases channel dimension and decreases spatial dimensions. We empirically show that such a spatial dimension reduction is beneficial to a transformer architecture as well, and propose a novel Pooling-based Vision Transformer (PiT) upon the original ViT model. We show that PiT achieves the improved model capability and generalization performance against ViT. Throughout the extensive experiments, we further show PiT outperforms the baseline on several tasks such as image classification, object detection, and robustness evaluation. Source codes and ImageNet models are available at https://github.com/naver-ai/pit

Vision Transformers (ViT), when paired with large-scale pretraining, have shown remarkable performance across various computer vision tasks, primarily due to their weak inductive bias. However, while such weak inductive bias aids in pretraining scalability, this may hinder the effective adaptation of ViTs for visuo-motor control tasks as a result of the absence of control-centric inductive biases. Such absent inductive biases include spatial locality and translation equivariance bias which convolutions naturally offer. To this end, we introduce Convolution Injector (CoIn), an add-on module that injects convolutions which are rich in locality and equivariance biases into a pretrained ViT for effective adaptation in visuo-motor control. We evaluate CoIn with three distinct types of pretrained ViTs (CLIP, MVP, VC-1) across 12 varied control tasks within three separate domains (Adroit, MetaWorld, DMC), and demonstrate that CoIn consistently enhances control task performance across all experimented environments and models, validating the effectiveness of providing pretrained ViTs with control-centric biases.

Large multimodal models excel in multimodal tasks but face significant computational challenges due to excessive computation on visual tokens. Unlike token reduction methods that focus on token-level redundancy, we identify and study the computation-level redundancy on vision tokens to ensure no information loss. Our key insight is that vision tokens from the pretrained vision encoder do not necessarily require all the heavy operations (e.g., self-attention, FFNs) in decoder-only LMMs and could be processed more lightly with proper designs. We designed a series of experiments to discover and progressively squeeze out the vision-related computation redundancy. Based on our findings, we propose ProxyV, a novel approach that utilizes proxy vision tokens to alleviate the computational burden on original vision tokens. ProxyV enhances efficiency without compromising performance and can even yield notable performance gains in scenarios with more moderate efficiency improvements. Furthermore, the flexibility of ProxyV is demonstrated through its combination with token reduction methods to boost efficiency further. The code will be made public at this https://github.com/penghao-wu/ProxyV URL.

The development of large language models (LLMs) has significantly advanced the emergence of large multimodal models (LMMs). While LMMs have achieved tremendous success by promoting the synergy between multimodal comprehension and creation, they often face challenges when confronted with out-of-distribution data. This is primarily due to their reliance on image encoders trained to encode images into task-relevant features, which may lead them to disregard irrelevant details. Delving into the modeling capabilities of diffusion models for images naturally prompts the question: Can diffusion models serve as the eyes of large language models for image perception? In this paper, we propose DEEM, a simple and effective approach that utilizes the generative feedback of diffusion models to align the semantic distributions of the image encoder. This addresses the drawbacks of previous methods that solely relied on image encoders like ViT, thereby enhancing the model's resilience against out-of-distribution samples and reducing visual hallucinations. Importantly, this is achieved without requiring additional training modules and with fewer training parameters. We extensively evaluated DEEM on both our newly constructed RobustVQA benchmark and another well-known benchmark, POPE, for object hallucination. Compared to the state-of-the-art interleaved content generation models, DEEM exhibits enhanced robustness and a superior capacity to alleviate model hallucinations while utilizing fewer trainable parameters, less pre-training data (10%), and a smaller base model size.

Attention-based vision models, such as Vision Transformer (ViT) and its variants, have shown promising performance in various computer vision tasks. However, these emerging architectures suffer from large model sizes and high computational costs, calling for efficient model compression solutions. To date, pruning ViTs has been well studied, while other compression strategies that have been widely applied in CNN compression, e.g., model factorization, is little explored in the context of ViT compression. This paper explores an efficient method for compressing vision transformers to enrich the toolset for obtaining compact attention-based vision models. Based on the new insight on the multi-head attention layer, we develop a highly efficient ViT compression solution, which outperforms the state-of-the-art pruning methods. For compressing DeiT-small and DeiT-base models on ImageNet, our proposed approach can achieve 0.45% and 0.76% higher top-1 accuracy even with fewer parameters. Our finding can also be applied to improve the customization efficiency of text-to-image diffusion models, with much faster training (up to 2.6times speedup) and lower extra storage cost (up to 1927.5times reduction) than the existing works.

This paper shows that masked autoencoders (MAE) are scalable self-supervised learners for computer vision. Our MAE approach is simple: we mask random patches of the input image and reconstruct the missing pixels. It is based on two core designs. First, we develop an asymmetric encoder-decoder architecture, with an encoder that operates only on the visible subset of patches (without mask tokens), along with a lightweight decoder that reconstructs the original image from the latent representation and mask tokens. Second, we find that masking a high proportion of the input image, e.g., 75%, yields a nontrivial and meaningful self-supervisory task. Coupling these two designs enables us to train large models efficiently and effectively: we accelerate training (by 3x or more) and improve accuracy. Our scalable approach allows for learning high-capacity models that generalize well: e.g., a vanilla ViT-Huge model achieves the best accuracy (87.8%) among methods that use only ImageNet-1K data. Transfer performance in downstream tasks outperforms supervised pre-training and shows promising scaling behavior.

Vision Transformer (ViT) models have recently emerged as powerful and versatile models for various visual tasks. Recently, a work called PMF has achieved promising results in few-shot image classification by utilizing pre-trained vision transformer models. However, PMF employs full fine-tuning for learning the downstream tasks, leading to significant overfitting and storage issues, especially in the remote sensing domain. In order to tackle these issues, we turn to the recently proposed parameter-efficient tuning methods, such as VPT, which updates only the newly added prompt parameters while keeping the pre-trained backbone frozen. Inspired by VPT, we propose the Meta Visual Prompt Tuning (MVP) method. Specifically, we integrate the VPT method into the meta-learning framework and tailor it to the remote sensing domain, resulting in an efficient framework for Few-Shot Remote Sensing Scene Classification (FS-RSSC). Furthermore, we introduce a novel data augmentation strategy based on patch embedding recombination to enhance the representation and diversity of scenes for classification purposes. Experiment results on the FS-RSSC benchmark demonstrate the superior performance of the proposed MVP over existing methods in various settings, such as various-way-various-shot, various-way-one-shot, and cross-domain adaptation.

Large-scale vision and language representation learning has shown promising improvements on various vision-language tasks. Most existing methods employ a transformer-based multimodal encoder to jointly model visual tokens (region-based image features) and word tokens. Because the visual tokens and word tokens are unaligned, it is challenging for the multimodal encoder to learn image-text interactions. In this paper, we introduce a contrastive loss to ALign the image and text representations BEfore Fusing (ALBEF) them through cross-modal attention, which enables more grounded vision and language representation learning. Unlike most existing methods, our method does not require bounding box annotations nor high-resolution images. In order to improve learning from noisy web data, we propose momentum distillation, a self-training method which learns from pseudo-targets produced by a momentum model. We provide a theoretical analysis of ALBEF from a mutual information maximization perspective, showing that different training tasks can be interpreted as different ways to generate views for an image-text pair. ALBEF achieves state-of-the-art performance on multiple downstream vision-language tasks. On image-text retrieval, ALBEF outperforms methods that are pre-trained on orders of magnitude larger datasets. On VQA and NLVR^2, ALBEF achieves absolute improvements of 2.37% and 3.84% compared to the state-of-the-art, while enjoying faster inference speed. Code and pre-trained models are available at https://github.com/salesforce/ALBEF/.

We investigate methods to reduce inference time and memory footprint in stable diffusion models by introducing lightweight decoders for both image and video synthesis. Traditional latent diffusion pipelines rely on large Variational Autoencoder decoders that can slow down generation and consume considerable GPU memory. We propose custom-trained decoders using lightweight Vision Transformer and Taming Transformer architectures. Experiments show up to 15% overall speed-ups for image generation on COCO2017 and up to 20 times faster decoding in the sub-module, with additional gains on UCF-101 for video tasks. Memory requirements are moderately reduced, and while there is a small drop in perceptual quality compared to the default decoder, the improvements in speed and scalability are crucial for large-scale inference scenarios such as generating 100K images. Our work is further contextualized by advances in efficient video generation, including dual masking strategies, illustrating a broader effort to improve the scalability and efficiency of generative models.

Vision Transformers (ViTs) have delivered remarkable progress through global self-attention, yet their quadratic complexity can become prohibitive for high-resolution inputs. In this work, we present ViT-Linearizer, a cross-architecture distillation framework that transfers rich ViT representations into a linear-time, recurrent-style model. Our approach leverages 1) activation matching, an intermediate constraint that encourages student to align its token-wise dependencies with those produced by the teacher, and 2) masked prediction, a contextual reconstruction objective that requires the student to predict the teacher's representations for unseen (masked) tokens, to effectively distill the quadratic self-attention knowledge into the student while maintaining efficient complexity. Empirically, our method provides notable speedups particularly for high-resolution tasks, significantly addressing the hardware challenges in inference. Additionally, it also elevates Mamba-based architectures' performance on standard vision benchmarks, achieving a competitive 84.3% top-1 accuracy on ImageNet with a base-sized model. Our results underscore the good potential of RNN-based solutions for large-scale visual tasks, bridging the gap between theoretical efficiency and real-world practice.

Current vision models typically maintain a fixed correspondence between their representation structure and image space. Each layer comprises a set of tokens arranged "on-the-grid," which biases patches or tokens to encode information at a specific spatio(-temporal) location. In this work we present Moving Off-the-Grid (MooG), a self-supervised video representation model that offers an alternative approach, allowing tokens to move "off-the-grid" to better enable them to represent scene elements consistently, even as they move across the image plane through time. By using a combination of cross-attention and positional embeddings we disentangle the representation structure and image structure. We find that a simple self-supervised objective--next frame prediction--trained on video data, results in a set of latent tokens which bind to specific scene structures and track them as they move. We demonstrate the usefulness of MooG's learned representation both qualitatively and quantitatively by training readouts on top of the learned representation on a variety of downstream tasks. We show that MooG can provide a strong foundation for different vision tasks when compared to "on-the-grid" baselines.

Large Foundation Models like Dust3r can produce high quality outputs such as pointmaps, camera intrinsics, and depth estimation, given stereo-image pairs as input. However, the application of these outputs on tasks like Visual Localization requires a large amount of inference time and compute resources. To address these limitations, in this paper, we propose the use of a knowledge distillation pipeline, where we aim to build a student-teacher model with Dust3r as the teacher and explore multiple architectures of student models that are trained using the 3D reconstructed points output by Dust3r. Our goal is to build student models that can learn scene-specific representations and output 3D points with replicable performance such as Dust3r. The data set we used to train our models is 12Scenes. We test two main architectures of models: a CNN-based architecture and a Vision Transformer based architecture. For each architecture, we also compare the use of pre-trained models against models built from scratch. We qualitatively compare the reconstructed 3D points output by the student model against Dust3r's and discuss the various features learned by the student model. We also perform ablation studies on the models through hyperparameter tuning. Overall, we observe that the Vision Transformer presents the best performance visually and quantitatively.

Convolutional neural networks (CNNs) have so far been the de-facto model for visual data. Recent work has shown that (Vision) Transformer models (ViT) can achieve comparable or even superior performance on image classification tasks. This raises a central question: how are Vision Transformers solving these tasks? Are they acting like convolutional networks, or learning entirely different visual representations? Analyzing the internal representation structure of ViTs and CNNs on image classification benchmarks, we find striking differences between the two architectures, such as ViT having more uniform representations across all layers. We explore how these differences arise, finding crucial roles played by self-attention, which enables early aggregation of global information, and ViT residual connections, which strongly propagate features from lower to higher layers. We study the ramifications for spatial localization, demonstrating ViTs successfully preserve input spatial information, with noticeable effects from different classification methods. Finally, we study the effect of (pretraining) dataset scale on intermediate features and transfer learning, and conclude with a discussion on connections to new architectures such as the MLP-Mixer.

Vision language models (VLMs) have seen growing adoption in recent years, but many still struggle with basic spatial reasoning errors. We hypothesize that this is due to VLMs adopting pre-trained vision backbones, specifically vision transformers (ViTs) trained with image-level supervision and minimal inductive biases. Such models may fail to encode the class contents at each position in the image, and our goal is to resolve this by ensuring that the vision backbone effectively captures both local and global image semantics. Our main insight is that we do not require new supervision to learn this capability -- pre-trained models contain significant knowledge of local semantics that we can extract and use for scalable self-supervision. We propose a new efficient post-training stage for ViTs called locality alignment and a novel fine-tuning procedure called MaskEmbed that uses a masked reconstruction loss to learn semantic contributions for each image patch. We first evaluate locality alignment with a vision-only benchmark, finding that it improves a model's performance at a patch-level semantic segmentation task, especially for strong backbones trained with image-caption pairs (e.g., CLIP and SigLIP). We then train a series of VLMs with and without locality alignment, and show that locality-aligned backbones improve performance across a range of benchmarks, particularly ones that involve spatial understanding (e.g., RefCOCO, OCID-Ref, TallyQA, VSR, AI2D). Overall, we demonstrate that we can efficiently learn local semantic extraction via a locality alignment stage, and that this procedure complements existing VLM training recipes that use off-the-shelf vision backbones.

Product retrieval is of great importance in the ecommerce domain. This paper introduces our 1st-place solution in eBay eProduct Visual Search Challenge (FGVC9), which is featured for an ensemble of about 20 models from vision models and vision-language models. While model ensemble is common, we show that combining the vision models and vision-language models brings particular benefits from their complementarity and is a key factor to our superiority. Specifically, for the vision models, we use a two-stage training pipeline which first learns from the coarse labels provided in the training set and then conducts fine-grained self-supervised training, yielding a coarse-to-fine metric learning manner. For the vision-language models, we use the textual description of the training image as the supervision signals for fine-tuning the image-encoder (feature extractor). With these designs, our solution achieves 0.7623 MAR@10, ranking the first place among all the competitors. The code is available at: https://github.com/WangWenhao0716/V2L{V^2L}.

High-resolution images offer more information about scenes that can improve model accuracy. However, the dominant model architecture in computer vision, the vision transformer (ViT), cannot effectively leverage larger images without finetuning -- ViTs poorly extrapolate to more patches at test time, although transformers offer sequence length flexibility. We attribute this shortcoming to the current patch position encoding methods, which create a distribution shift when extrapolating. We propose a drop-in replacement for the position encoding of plain ViTs that restricts attention heads to fixed fields of view, pointed in different directions, using 2D attention masks. Our novel method, called LookHere, provides translation-equivariance, ensures attention head diversity, and limits the distribution shift that attention heads face when extrapolating. We demonstrate that LookHere improves performance on classification (avg. 1.6%), against adversarial attack (avg. 5.4%), and decreases calibration error (avg. 1.5%) -- on ImageNet without extrapolation. With extrapolation, LookHere outperforms the current SoTA position encoding method, 2D-RoPE, by 21.7% on ImageNet when trained at 224^2 px and tested at 1024^2 px. Additionally, we release a high-resolution test set to improve the evaluation of high-resolution image classifiers, called ImageNet-HR.

Resource-constrained hardware, such as edge devices or cell phones, often rely on cloud servers to provide the required computational resources for inference in deep vision models. However, transferring image and video data from an edge or mobile device to a cloud server requires coding to deal with network constraints. The use of standardized codecs, such as JPEG or H.264, is prevalent and required to ensure interoperability. This paper aims to examine the implications of employing standardized codecs within deep vision pipelines. We find that using JPEG and H.264 coding significantly deteriorates the accuracy across a broad range of vision tasks and models. For instance, strong compression rates reduce semantic segmentation accuracy by more than 80% in mIoU. In contrast to previous findings, our analysis extends beyond image and action classification to localization and dense prediction tasks, thus providing a more comprehensive perspective.

Vision Transformers (ViTs) have significantly advanced image classification by applying self-attention on patch embeddings. However, the standard MLP blocks in each Transformer layer may not capture complex nonlinear dependencies optimally. In this paper, we propose ViKANformer, a Vision Transformer where we replace the MLP sub-layers with Kolmogorov-Arnold Network (KAN) expansions, including Vanilla KAN, Efficient-KAN, Fast-KAN, SineKAN, and FourierKAN, while also examining a Flash Attention variant. By leveraging the Kolmogorov-Arnold theorem, which guarantees that multivariate continuous functions can be expressed via sums of univariate continuous functions, we aim to boost representational power. Experimental results on MNIST demonstrate that SineKAN, Fast-KAN, and a well-tuned Vanilla KAN can achieve over 97% accuracy, albeit with increased training overhead. This trade-off highlights that KAN expansions may be beneficial if computational cost is acceptable. We detail the expansions, present training/test accuracy and F1/ROC metrics, and provide pseudocode and hyperparameters for reproducibility. Finally, we compare ViKANformer to a simple MLP and a small CNN baseline on MNIST, illustrating the efficiency of Transformer-based methods even on a small-scale dataset.

Most recent semantic segmentation methods adopt a fully-convolutional network (FCN) with an encoder-decoder architecture. The encoder progressively reduces the spatial resolution and learns more abstract/semantic visual concepts with larger receptive fields. Since context modeling is critical for segmentation, the latest efforts have been focused on increasing the receptive field, through either dilated/atrous convolutions or inserting attention modules. However, the encoder-decoder based FCN architecture remains unchanged. In this paper, we aim to provide an alternative perspective by treating semantic segmentation as a sequence-to-sequence prediction task. Specifically, we deploy a pure transformer (ie, without convolution and resolution reduction) to encode an image as a sequence of patches. With the global context modeled in every layer of the transformer, this encoder can be combined with a simple decoder to provide a powerful segmentation model, termed SEgmentation TRansformer (SETR). Extensive experiments show that SETR achieves new state of the art on ADE20K (50.28% mIoU), Pascal Context (55.83% mIoU) and competitive results on Cityscapes. Particularly, we achieve the first position in the highly competitive ADE20K test server leaderboard on the day of submission.

We present an unsupervised representation learning approach that compactly encodes the motion dependencies in videos. Given a pair of images from a video clip, our framework learns to predict the long-term 3D motions. To reduce the complexity of the learning framework, we propose to describe the motion as a sequence of atomic 3D flows computed with RGB-D modality. We use a Recurrent Neural Network based Encoder-Decoder framework to predict these sequences of flows. We argue that in order for the decoder to reconstruct these sequences, the encoder must learn a robust video representation that captures long-term motion dependencies and spatial-temporal relations. We demonstrate the effectiveness of our learned temporal representations on activity classification across multiple modalities and datasets such as NTU RGB+D and MSR Daily Activity 3D. Our framework is generic to any input modality, i.e., RGB, Depth, and RGB-D videos.

Self-supervised representation learning approaches have recently surpassed their supervised learning counterparts on downstream tasks like object detection and image classification. Somewhat mysteriously the recent gains in performance come from training instance classification models, treating each image and it's augmented versions as samples of a single class. In this work, we first present quantitative experiments to demystify these gains. We demonstrate that approaches like MOCO and PIRL learn occlusion-invariant representations. However, they fail to capture viewpoint and category instance invariance which are crucial components for object recognition. Second, we demonstrate that these approaches obtain further gains from access to a clean object-centric training dataset like Imagenet. Finally, we propose an approach to leverage unstructured videos to learn representations that possess higher viewpoint invariance. Our results show that the learned representations outperform MOCOv2 trained on the same data in terms of invariances encoded and the performance on downstream image classification and semantic segmentation tasks.

The goal of this work is to characterize the representational impact that foveation operations have for machine vision systems, inspired by the foveated human visual system, which has higher acuity at the center of gaze and texture-like encoding in the periphery. To do so, we introduce models consisting of a first-stage fixed image transform followed by a second-stage learnable convolutional neural network, and we varied the first stage component. The primary model has a foveated-textural input stage, which we compare to a model with foveated-blurred input and a model with spatially-uniform blurred input (both matched for perceptual compression), and a final reference model with minimal input-based compression. We find that: 1) the foveated-texture model shows similar scene classification accuracy as the reference model despite its compressed input, with greater i.i.d. generalization than the other models; 2) the foveated-texture model has greater sensitivity to high-spatial frequency information and greater robustness to occlusion, w.r.t the comparison models; 3) both the foveated systems, show a stronger center image-bias relative to the spatially-uniform systems even with a weight sharing constraint. Critically, these results are preserved over different classical CNN architectures throughout their learning dynamics. Altogether, this suggests that foveation with peripheral texture-based computations yields an efficient, distinct, and robust representational format of scene information, and provides symbiotic computational insight into the representational consequences that texture-based peripheral encoding may have for processing in the human visual system, while also potentially inspiring the next generation of computer vision models via spatially-adaptive computation. Code + Data available here: https://github.com/ArturoDeza/EmergentProperties

Visual speech recognition models extract visual features in a hierarchical manner. At the lower level, there is a visual front-end with a limited temporal receptive field that processes the raw pixels depicting the lips or faces. At the higher level, there is an encoder that attends to the embeddings produced by the front-end over a large temporal receptive field. Previous work has focused on improving the visual front-end of the model to extract more useful features for speech recognition. Surprisingly, our work shows that complex visual front-ends are not necessary. Instead of allocating resources to a sophisticated visual front-end, we find that a linear visual front-end paired with a larger Conformer encoder results in lower latency, more efficient memory usage, and improved WER performance. We achieve a new state-of-the-art of 12.8% WER for visual speech recognition on the TED LRS3 dataset, which rivals the performance of audio-only models from just four years ago.

Vision Transformers (ViTs) have gained significant popularity in recent years and have proliferated into many applications. However, their behavior under different learning paradigms is not well explored. We compare ViTs trained through different methods of supervision, and show that they learn a diverse range of behaviors in terms of their attention, representations, and downstream performance. We also discover ViT behaviors that are consistent across supervision, including the emergence of Offset Local Attention Heads. These are self-attention heads that attend to a token adjacent to the current token with a fixed directional offset, a phenomenon that to the best of our knowledge has not been highlighted in any prior work. Our analysis shows that ViTs are highly flexible and learn to process local and global information in different orders depending on their training method. We find that contrastive self-supervised methods learn features that are competitive with explicitly supervised features, and they can even be superior for part-level tasks. We also find that the representations of reconstruction-based models show non-trivial similarity to contrastive self-supervised models. Project website (https://www.cs.umd.edu/~sakshams/vit_analysis) and code (https://www.github.com/mwalmer-umd/vit_analysis) are publicly available.

Despite the great success of Large Vision-Language Models (LVLMs), they inevitably suffer from hallucination. As we know, both the visual encoder and the Large Language Model (LLM) decoder in LVLMs are Transformer-based, allowing the model to extract visual information and generate text outputs via attention mechanisms. We find that the attention distribution of LLM decoder on image tokens is highly consistent with the visual encoder and both distributions tend to focus on particular background tokens rather than the referred objects in the image. We attribute to the unexpected attention distribution to an inherent flaw in the visual encoder itself, which misguides LLMs to over emphasize the redundant information and generate object hallucination. To address the issue, we propose DAMRO, a novel training-free strategy that Dive into Attention Mechanism of LVLM to Reduce Object Hallucination. Specifically, our approach employs classification token (CLS) of ViT to filter out high-attention outlier tokens scattered in the background and then eliminate their influence during decoding stage. We evaluate our method on LVLMs including LLaVA-1.5, LLaVA-NeXT and InstructBLIP, using various benchmarks such as POPE, CHAIR, MME and GPT-4V Aided Evaluation. The results demonstrate that our approach significantly reduces the impact of these outlier tokens, thus effectively alleviating the hallucination of LVLMs. The code of our method will be released soon.

We present DietNeRF, a 3D neural scene representation estimated from a few images. Neural Radiance Fields (NeRF) learn a continuous volumetric representation of a scene through multi-view consistency, and can be rendered from novel viewpoints by ray casting. While NeRF has an impressive ability to reconstruct geometry and fine details given many images, up to 100 for challenging 360{\deg} scenes, it often finds a degenerate solution to its image reconstruction objective when only a few input views are available. To improve few-shot quality, we propose DietNeRF. We introduce an auxiliary semantic consistency loss that encourages realistic renderings at novel poses. DietNeRF is trained on individual scenes to (1) correctly render given input views from the same pose, and (2) match high-level semantic attributes across different, random poses. Our semantic loss allows us to supervise DietNeRF from arbitrary poses. We extract these semantics using a pre-trained visual encoder such as CLIP, a Vision Transformer trained on hundreds of millions of diverse single-view, 2D photographs mined from the web with natural language supervision. In experiments, DietNeRF improves the perceptual quality of few-shot view synthesis when learned from scratch, can render novel views with as few as one observed image when pre-trained on a multi-view dataset, and produces plausible completions of completely unobserved regions.

This study explores the potential of open-source video conditional generation models as encoders for downstream tasks, focusing on instance segmentation using the BAIR Robot Pushing Dataset. The researchers propose using video prediction models as general visual encoders, leveraging their ability to capture critical spatial and temporal information which is essential for tasks such as instance segmentation. Inspired by human vision studies, particularly Gestalts principle of common fate, the approach aims to develop a latent space representative of motion from images to effectively discern foreground from background information. The researchers utilize a 3D Vector-Quantized Variational Autoencoder 3D VQVAE video generative encoder model conditioned on an input frame, coupled with downstream segmentation tasks. Experiments involve adapting pre-trained video generative models, analyzing their latent spaces, and training custom decoders for foreground-background segmentation. The findings demonstrate promising results in leveraging generative pretext learning for downstream tasks, working towards enhanced scene analysis and segmentation in computer vision applications.

After their initial success in natural language processing, transformer architectures have rapidly gained traction in computer vision, providing state-of-the-art results for tasks such as image classification, detection, segmentation, and video analysis. We offer three insights based on simple and easy to implement variants of vision transformers. (1) The residual layers of vision transformers, which are usually processed sequentially, can to some extent be processed efficiently in parallel without noticeably affecting the accuracy. (2) Fine-tuning the weights of the attention layers is sufficient to adapt vision transformers to a higher resolution and to other classification tasks. This saves compute, reduces the peak memory consumption at fine-tuning time, and allows sharing the majority of weights across tasks. (3) Adding MLP-based patch pre-processing layers improves Bert-like self-supervised training based on patch masking. We evaluate the impact of these design choices using the ImageNet-1k dataset, and confirm our findings on the ImageNet-v2 test set. Transfer performance is measured across six smaller datasets.

We present SOLO, a single transformer for Scalable visiOn-Language mOdeling. Current large vision-language models (LVLMs) such as LLaVA mostly employ heterogeneous architectures that connect pre-trained visual encoders with large language models (LLMs) to facilitate visual recognition and complex reasoning. Although achieving remarkable performance with relatively lightweight training, we identify four primary scalability limitations: (1) The visual capacity is constrained by pre-trained visual encoders, which are typically an order of magnitude smaller than LLMs. (2) The heterogeneous architecture complicates the use of established hardware and software infrastructure. (3) Study of scaling laws on such architecture must consider three separate components - visual encoder, connector, and LLMs, which complicates the analysis. (4) The use of existing visual encoders typically requires following a pre-defined specification of image inputs pre-processing, for example, by reshaping inputs to fixed-resolution square images, which presents difficulties in processing and training on high-resolution images or those with unusual aspect ratio. A unified single Transformer architecture, like SOLO, effectively addresses these scalability concerns in LVLMs; however, its limited adoption in the modern context likely stems from the absence of reliable training recipes that balance both modalities and ensure stable training for billion-scale models. In this paper, we introduce the first open-source training recipe for developing SOLO, an open-source 7B LVLM using moderate academic resources. The training recipe involves initializing from LLMs, sequential pre-training on ImageNet and web-scale data, and instruction fine-tuning on our curated high-quality datasets. On extensive evaluation, SOLO demonstrates performance comparable to LLaVA-v1.5-7B, particularly excelling in visual mathematical reasoning.

Vision transformers (ViTs) are quickly becoming the de-facto architecture for computer vision, yet we understand very little about why they work and what they learn. While existing studies visually analyze the mechanisms of convolutional neural networks, an analogous exploration of ViTs remains challenging. In this paper, we first address the obstacles to performing visualizations on ViTs. Assisted by these solutions, we observe that neurons in ViTs trained with language model supervision (e.g., CLIP) are activated by semantic concepts rather than visual features. We also explore the underlying differences between ViTs and CNNs, and we find that transformers detect image background features, just like their convolutional counterparts, but their predictions depend far less on high-frequency information. On the other hand, both architecture types behave similarly in the way features progress from abstract patterns in early layers to concrete objects in late layers. In addition, we show that ViTs maintain spatial information in all layers except the final layer. In contrast to previous works, we show that the last layer most likely discards the spatial information and behaves as a learned global pooling operation. Finally, we conduct large-scale visualizations on a wide range of ViT variants, including DeiT, CoaT, ConViT, PiT, Swin, and Twin, to validate the effectiveness of our method.

Current machine learning models for vision are often highly specialized and limited to a single modality and task. In contrast, recent large language models exhibit a wide range of capabilities, hinting at a possibility for similarly versatile models in computer vision. In this paper, we take a step in this direction and propose a multimodal training scheme called 4M. It consists of training a single unified Transformer encoder-decoder using a masked modeling objective across a wide range of input/output modalities - including text, images, geometric, and semantic modalities, as well as neural network feature maps. 4M achieves scalability by unifying the representation space of all modalities through mapping them into discrete tokens and performing multimodal masked modeling on a small randomized subset of tokens. 4M leads to models that exhibit several key capabilities: (1) they can perform a diverse set of vision tasks out of the box, (2) they excel when fine-tuned for unseen downstream tasks or new input modalities, and (3) they can function as a generative model that can be conditioned on arbitrary modalities, enabling a wide variety of expressive multimodal editing capabilities with remarkable flexibility. Through experimental analyses, we demonstrate the potential of 4M for training versatile and scalable foundation models for vision tasks, setting the stage for further exploration in multimodal learning for vision and other domains.

Vision transformers (ViTs) have gained popularity recently. Even without customized image operators such as convolutions, ViTs can yield competitive performance when properly trained on massive data. However, the computational overhead of ViTs remains prohibitive, due to stacking multi-head self-attention modules and else. Compared to the vast literature and prevailing success in compressing convolutional neural networks, the study of Vision Transformer compression has also just emerged, and existing works focused on one or two aspects of compression. This paper proposes a unified ViT compression framework that seamlessly assembles three effective techniques: pruning, layer skipping, and knowledge distillation. We formulate a budget-constrained, end-to-end optimization framework, targeting jointly learning model weights, layer-wise pruning ratios/masks, and skip configurations, under a distillation loss. The optimization problem is then solved using the primal-dual algorithm. Experiments are conducted with several ViT variants, e.g. DeiT and T2T-ViT backbones on the ImageNet dataset, and our approach consistently outperforms recent competitors. For example, DeiT-Tiny can be trimmed down to 50\% of the original FLOPs almost without losing accuracy. Codes are available online:~https://github.com/VITA-Group/UVC.

Self-Supervised Learning (SSL) has become a prominent approach for acquiring visual representations across various tasks, yet its application in fine-grained visual recognition (FGVR) is challenged by the intricate task of distinguishing subtle differences between categories. To overcome this, we introduce an novel strategy that boosts SSL's ability to extract critical discriminative features vital for FGVR. This approach creates synthesized data pairs to guide the model to focus on discriminative features critical for FGVR during SSL. We start by identifying non-discriminative features using two main criteria: features with low variance that fail to effectively separate data and those deemed less important by Grad-CAM induced from the SSL loss. We then introduce perturbations to these non-discriminative features while preserving discriminative ones. A decoder is employed to reconstruct images from both perturbed and original feature vectors to create data pairs. An encoder is trained on such generated data pairs to become invariant to variations in non-discriminative dimensions while focusing on discriminative features, thereby improving the model's performance in FGVR tasks. We demonstrate the promising FGVR performance of the proposed approach through extensive evaluation on a wide variety of datasets.

This paper reveals that large language models (LLMs), despite being trained solely on textual data, are surprisingly strong encoders for purely visual tasks in the absence of language. Even more intriguingly, this can be achieved by a simple yet previously overlooked strategy -- employing a frozen transformer block from pre-trained LLMs as a constituent encoder layer to directly process visual tokens. Our work pushes the boundaries of leveraging LLMs for computer vision tasks, significantly departing from conventional practices that typically necessitate a multi-modal vision-language setup with associated language prompts, inputs, or outputs. We demonstrate that our approach consistently enhances performance across a diverse range of tasks, encompassing pure 2D and 3D visual recognition tasks (e.g., image and point cloud classification), temporal modeling tasks (e.g., action recognition), non-semantic tasks (e.g., motion forecasting), and multi-modal tasks (e.g., 2D/3D visual question answering and image-text retrieval). Such improvements are a general phenomenon, applicable to various types of LLMs (e.g., LLaMA and OPT) and different LLM transformer blocks. We additionally propose the information filtering hypothesis to explain the effectiveness of pre-trained LLMs in visual encoding -- the pre-trained LLM transformer blocks discern informative visual tokens and further amplify their effect. This hypothesis is empirically supported by the observation that the feature activation, after training with LLM transformer blocks, exhibits a stronger focus on relevant regions. We hope that our work inspires new perspectives on utilizing LLMs and deepening our understanding of their underlying mechanisms. Code is available at https://github.com/ziqipang/LM4VisualEncoding.

We present a simple approach which can turn a ViT encoder into an efficient video model, which can seamlessly work with both image and video inputs. By sparsely sampling the inputs, the model is able to do training and inference from both inputs. The model is easily scalable and can be adapted to large-scale pre-trained ViTs without requiring full finetuning. The model achieves SOTA results and the code will be open-sourced.

Human visual recognition is a sparse process, where only a few salient visual cues are attended to rather than traversing every detail uniformly. However, most current vision networks follow a dense paradigm, processing every single visual unit (e.g,, pixel or patch) in a uniform manner. In this paper, we challenge this dense paradigm and present a new method, coined SparseFormer, to imitate human's sparse visual recognition in an end-to-end manner. SparseFormer learns to represent images using a highly limited number of tokens (down to 49) in the latent space with sparse feature sampling procedure instead of processing dense units in the original pixel space. Therefore, SparseFormer circumvents most of dense operations on the image space and has much lower computational costs. Experiments on the ImageNet classification benchmark dataset show that SparseFormer achieves performance on par with canonical or well-established models while offering better accuracy-throughput tradeoff. Moreover, the design of our network can be easily extended to the video classification with promising performance at lower computational costs. We hope that our work can provide an alternative way for visual modeling and inspire further research on sparse neural architectures. The code will be publicly available at https://github.com/showlab/sparseformer

To truly understand vision models, we must not only interpret their learned features but also validate these interpretations through controlled experiments. Current approaches either provide interpretable features without the ability to test their causal influence, or enable model editing without interpretable controls. We present a unified framework using sparse autoencoders (SAEs) that bridges this gap, allowing us to discover human-interpretable visual features and precisely manipulate them to test hypotheses about model behavior. By applying our method to state-of-the-art vision models, we reveal key differences in the semantic abstractions learned by models with different pre-training objectives. We then demonstrate the practical usage of our framework through controlled interventions across multiple vision tasks. We show that SAEs can reliably identify and manipulate interpretable visual features without model re-training, providing a powerful tool for understanding and controlling vision model behavior. We provide code, demos and models on our project website: https://osu-nlp-group.github.io/SAE-V.

Predicting salient regions in natural images requires the detection of objects that are present in a scene. To develop robust representations for this challenging task, high-level visual features at multiple spatial scales must be extracted and augmented with contextual information. However, existing models aimed at explaining human fixation maps do not incorporate such a mechanism explicitly. Here we propose an approach based on a convolutional neural network pre-trained on a large-scale image classification task. The architecture forms an encoder-decoder structure and includes a module with multiple convolutional layers at different dilation rates to capture multi-scale features in parallel. Moreover, we combine the resulting representations with global scene information for accurately predicting visual saliency. Our model achieves competitive and consistent results across multiple evaluation metrics on two public saliency benchmarks and we demonstrate the effectiveness of the suggested approach on five datasets and selected examples. Compared to state of the art approaches, the network is based on a lightweight image classification backbone and hence presents a suitable choice for applications with limited computational resources, such as (virtual) robotic systems, to estimate human fixations across complex natural scenes.

We launch EVA, a vision-centric foundation model to explore the limits of visual representation at scale using only publicly accessible data. EVA is a vanilla ViT pre-trained to reconstruct the masked out image-text aligned vision features conditioned on visible image patches. Via this pretext task, we can efficiently scale up EVA to one billion parameters, and sets new records on a broad range of representative vision downstream tasks, such as image recognition, video action recognition, object detection, instance segmentation and semantic segmentation without heavy supervised training. Moreover, we observe quantitative changes in scaling EVA result in qualitative changes in transfer learning performance that are not present in other models. For instance, EVA takes a great leap in the challenging large vocabulary instance segmentation task: our model achieves almost the same state-of-the-art performance on LVISv1.0 dataset with over a thousand categories and COCO dataset with only eighty categories. Beyond a pure vision encoder, EVA can also serve as a vision-centric, multi-modal pivot to connect images and text. We find initializing the vision tower of a giant CLIP from EVA can greatly stabilize the training and outperform the training from scratch counterpart with much fewer samples and less compute, providing a new direction for scaling up and accelerating the costly training of multi-modal foundation models. To facilitate future research, we release all the code and models at https://github.com/baaivision/EVA.

Existing object detection frameworks are usually built on a single format of object/part representation, i.e., anchor/proposal rectangle boxes in RetinaNet and Faster R-CNN, center points in FCOS and RepPoints, and corner points in CornerNet. While these different representations usually drive the frameworks to perform well in different aspects, e.g., better classification or finer localization, it is in general difficult to combine these representations in a single framework to make good use of each strength, due to the heterogeneous or non-grid feature extraction by different representations. This paper presents an attention-based decoder module similar as that in Transformer~vaswani2017attention to bridge other representations into a typical object detector built on a single representation format, in an end-to-end fashion. The other representations act as a set of key instances to strengthen the main query representation features in the vanilla detectors. Novel techniques are proposed towards efficient computation of the decoder module, including a key sampling approach and a shared location embedding approach. The proposed module is named bridging visual representations (BVR). It can perform in-place and we demonstrate its broad effectiveness in bridging other representations into prevalent object detection frameworks, including RetinaNet, Faster R-CNN, FCOS and ATSS, where about 1.5sim3.0 AP improvements are achieved. In particular, we improve a state-of-the-art framework with a strong backbone by about 2.0 AP, reaching 52.7 AP on COCO test-dev. The resulting network is named RelationNet++. The code will be available at https://github.com/microsoft/RelationNet2.

Standardized lossy video coding is at the core of almost all real-world video processing pipelines. Rate control is used to enable standard codecs to adapt to different network bandwidth conditions or storage constraints. However, standard video codecs (e.g., H.264) and their rate control modules aim to minimize video distortion w.r.t. human quality assessment. We demonstrate empirically that standard-coded videos vastly deteriorate the performance of deep vision models. To overcome the deterioration of vision performance, this paper presents the first end-to-end learnable deep video codec control that considers both bandwidth constraints and downstream deep vision performance, while adhering to existing standardization. We demonstrate that our approach better preserves downstream deep vision performance than traditional standard video coding.

We investigate the CLIP image encoder by analyzing how individual model components affect the final representation. We decompose the image representation as a sum across individual image patches, model layers, and attention heads, and use CLIP's text representation to interpret the summands. Interpreting the attention heads, we characterize each head's role by automatically finding text representations that span its output space, which reveals property-specific roles for many heads (e.g. location or shape). Next, interpreting the image patches, we uncover an emergent spatial localization within CLIP. Finally, we use this understanding to remove spurious features from CLIP and to create a strong zero-shot image segmenter. Our results indicate that a scalable understanding of transformer models is attainable and can be used to repair and improve models.

The problem of visual metamerism is defined as finding a family of perceptually indistinguishable, yet physically different images. In this paper, we propose our NeuroFovea metamer model, a foveated generative model that is based on a mixture of peripheral representations and style transfer forward-pass algorithms. Our gradient-descent free model is parametrized by a foveated VGG19 encoder-decoder which allows us to encode images in high dimensional space and interpolate between the content and texture information with adaptive instance normalization anywhere in the visual field. Our contributions include: 1) A framework for computing metamers that resembles a noisy communication system via a foveated feed-forward encoder-decoder network -- We observe that metamerism arises as a byproduct of noisy perturbations that partially lie in the perceptual null space; 2) A perceptual optimization scheme as a solution to the hyperparametric nature of our metamer model that requires tuning of the image-texture tradeoff coefficients everywhere in the visual field which are a consequence of internal noise; 3) An ABX psychophysical evaluation of our metamers where we also find that the rate of growth of the receptive fields in our model match V1 for reference metamers and V2 between synthesized samples. Our model also renders metamers at roughly a second, presenting a times1000 speed-up compared to the previous work, which allows for tractable data-driven metamer experiments.

We present VicaSplat, a novel framework for joint 3D Gaussians reconstruction and camera pose estimation from a sequence of unposed video frames, which is a critical yet underexplored task in real-world 3D applications. The core of our method lies in a novel transformer-based network architecture. In particular, our model starts with an image encoder that maps each image to a list of visual tokens. All visual tokens are concatenated with additional inserted learnable camera tokens. The obtained tokens then fully communicate with each other within a tailored transformer decoder. The camera tokens causally aggregate features from visual tokens of different views, and further modulate them frame-wisely to inject view-dependent features. 3D Gaussian splats and camera pose parameters can then be estimated via different prediction heads. Experiments show that VicaSplat surpasses baseline methods for multi-view inputs, and achieves comparable performance to prior two-view approaches. Remarkably, VicaSplat also demonstrates exceptional cross-dataset generalization capability on the ScanNet benchmark, achieving superior performance without any fine-tuning. Project page: https://lizhiqi49.github.io/VicaSplat.

In this work we propose a novel joint training method for Visual Place Recognition (VPR), which simultaneously learns a global descriptor and a pair classifier for re-ranking. The pair classifier can predict whether a given pair of images are from the same place or not. The network only comprises Vision Transformer components for both the encoder and the pair classifier, and both components are trained using their respective class tokens. In existing VPR methods, typically the network is initialized using pre-trained weights from a generic image dataset such as ImageNet. In this work we propose an alternative pre-training strategy, by using Siamese Masked Image Modelling as a pre-training task. We propose a Place-aware image sampling procedure from a collection of large VPR datasets for pre-training our model, to learn visual features tuned specifically for VPR. By re-using the Mask Image Modelling encoder and decoder weights in the second stage of training, Pair-VPR can achieve state-of-the-art VPR performance across five benchmark datasets with a ViT-B encoder, along with further improvements in localization recall with larger encoders. The Pair-VPR website is: https://csiro-robotics.github.io/Pair-VPR.

Recent progress in vision Transformers exhibits great success in various tasks driven by the new spatial modeling mechanism based on dot-product self-attention. In this paper, we show that the key ingredients behind the vision Transformers, namely input-adaptive, long-range and high-order spatial interactions, can also be efficiently implemented with a convolution-based framework. We present the Recursive Gated Convolution (g^nConv) that performs high-order spatial interactions with gated convolutions and recursive designs. The new operation is highly flexible and customizable, which is compatible with various variants of convolution and extends the two-order interactions in self-attention to arbitrary orders without introducing significant extra computation. g^nConv can serve as a plug-and-play module to improve various vision Transformers and convolution-based models. Based on the operation, we construct a new family of generic vision backbones named HorNet. Extensive experiments on ImageNet classification, COCO object detection and ADE20K semantic segmentation show HorNet outperform Swin Transformers and ConvNeXt by a significant margin with similar overall architecture and training configurations. HorNet also shows favorable scalability to more training data and larger model sizes. Apart from the effectiveness in visual encoders, we also show g^nConv can be applied to task-specific decoders and consistently improve dense prediction performance with less computation. Our results demonstrate that g^nConv can be a new basic module for visual modeling that effectively combines the merits of both vision Transformers and CNNs. Code is available at https://github.com/raoyongming/HorNet

Deep learning models are increasingly deployed on resource-constrained edge devices for real-time data analytics. In recent years, Vision Transformer models and their variants have demonstrated outstanding performance across various computer vision tasks. However, their high computational demands and inference latency pose significant challenges for model deployment on resource-constraint edge devices. To address this issue, we propose a novel Vision Transformer splitting framework, ED-ViT, designed to execute complex models across multiple edge devices efficiently. Specifically, we partition Vision Transformer models into several sub-models, where each sub-model is tailored to handle a specific subset of data classes. To further minimize computation overhead and inference latency, we introduce a class-wise pruning technique that reduces the size of each sub-model. We conduct extensive experiments on five datasets with three model structures, demonstrating that our approach significantly reduces inference latency on edge devices and achieves a model size reduction of up to 28.9 times and 34.1 times, respectively, while maintaining test accuracy comparable to the original Vision Transformer. Additionally, we compare ED-ViT with two state-of-the-art methods that deploy CNN and SNN models on edge devices, evaluating accuracy, inference time, and overall model size. Our comprehensive evaluation underscores the effectiveness of the proposed ED-ViT framework.

Estimating accurate depth from a single image is challenging because it is an ill-posed problem as infinitely many 3D scenes can be projected to the same 2D scene. However, recent works based on deep convolutional neural networks show great progress with plausible results. The convolutional neural networks are generally composed of two parts: an encoder for dense feature extraction and a decoder for predicting the desired depth. In the encoder-decoder schemes, repeated strided convolution and spatial pooling layers lower the spatial resolution of transitional outputs, and several techniques such as skip connections or multi-layer deconvolutional networks are adopted to recover the original resolution for effective dense prediction. In this paper, for more effective guidance of densely encoded features to the desired depth prediction, we propose a network architecture that utilizes novel local planar guidance layers located at multiple stages in the decoding phase. We show that the proposed method outperforms the state-of-the-art works with significant margin evaluating on challenging benchmarks. We also provide results from an ablation study to validate the effectiveness of the proposed method.

We propose pixelNeRF, a learning framework that predicts a continuous neural scene representation conditioned on one or few input images. The existing approach for constructing neural radiance fields involves optimizing the representation to every scene independently, requiring many calibrated views and significant compute time. We take a step towards resolving these shortcomings by introducing an architecture that conditions a NeRF on image inputs in a fully convolutional manner. This allows the network to be trained across multiple scenes to learn a scene prior, enabling it to perform novel view synthesis in a feed-forward manner from a sparse set of views (as few as one). Leveraging the volume rendering approach of NeRF, our model can be trained directly from images with no explicit 3D supervision. We conduct extensive experiments on ShapeNet benchmarks for single image novel view synthesis tasks with held-out objects as well as entire unseen categories. We further demonstrate the flexibility of pixelNeRF by demonstrating it on multi-object ShapeNet scenes and real scenes from the DTU dataset. In all cases, pixelNeRF outperforms current state-of-the-art baselines for novel view synthesis and single image 3D reconstruction. For the video and code, please visit the project website: https://alexyu.net/pixelnerf

Recent advances in Vision Transformer (ViT) have demonstrated its impressive performance in image classification, which makes it a promising alternative to Convolutional Neural Network (CNN). Unlike CNNs, ViT represents an input image as a sequence of image patches. The patch-based input image representation makes the following question interesting: How does ViT perform when individual input image patches are perturbed with natural corruptions or adversarial perturbations, compared to CNNs? In this work, we study the robustness of ViT to patch-wise perturbations. Surprisingly, we find that ViTs are more robust to naturally corrupted patches than CNNs, whereas they are more vulnerable to adversarial patches. Furthermore, we discover that the attention mechanism greatly affects the robustness of vision transformers. Specifically, the attention module can help improve the robustness of ViT by effectively ignoring natural corrupted patches. However, when ViTs are attacked by an adversary, the attention mechanism can be easily fooled to focus more on the adversarially perturbed patches and cause a mistake. Based on our analysis, we propose a simple temperature-scaling based method to improve the robustness of ViT against adversarial patches. Extensive qualitative and quantitative experiments are performed to support our findings, understanding, and improvement of ViT robustness to patch-wise perturbations across a set of transformer-based architectures.

We study neural image compression based on the Sparse Visual Representation (SVR), where images are embedded into a discrete latent space spanned by learned visual codebooks. By sharing codebooks with the decoder, the encoder transfers integer codeword indices that are efficient and cross-platform robust, and the decoder retrieves the embedded latent feature using the indices for reconstruction. Previous SVR-based compression lacks effective mechanism for rate-distortion tradeoffs, where one can only pursue either high reconstruction quality or low transmission bitrate. We propose a Masked Adaptive Codebook learning (M-AdaCode) method that applies masks to the latent feature subspace to balance bitrate and reconstruction quality. A set of semantic-class-dependent basis codebooks are learned, which are weighted combined to generate a rich latent feature for high-quality reconstruction. The combining weights are adaptively derived from each input image, providing fidelity information with additional transmission costs. By masking out unimportant weights in the encoder and recovering them in the decoder, we can trade off reconstruction quality for transmission bits, and the masking rate controls the balance between bitrate and distortion. Experiments over the standard JPEG-AI dataset demonstrate the effectiveness of our M-AdaCode approach.

Transformers combined with convolutional encoders have been recently used for hand gesture recognition (HGR) using micro-Doppler signatures. We propose a vision-transformer-based architecture for HGR with multi-antenna continuous-wave Doppler radar receivers. The proposed architecture consists of three modules: a convolutional encoderdecoder, an attention module with three transformer layers, and a multi-layer perceptron. The novel convolutional decoder helps to feed patches with larger sizes to the attention module for improved feature extraction. Experimental results obtained with a dataset corresponding to a two-antenna continuous-wave Doppler radar receiver operating at 24 GHz (published by Skaria et al.) confirm that the proposed architecture achieves an accuracy of 98.3% which substantially surpasses the state-of-the-art on the used dataset.

Transformers are widely used for solving tasks in natural language processing, computer vision, speech, and music domains. In this paper, we talk about the efficiency of transformers in terms of memory (the number of parameters), computation cost (number of floating points operations), and performance of models, including accuracy, the robustness of the model, and fair \& bias-free features. We mainly discuss the vision transformer for the image classification task. Our contribution is to introduce an efficient 360 framework, which includes various aspects of the vision transformer, to make it more efficient for industrial applications. By considering those applications, we categorize them into multiple dimensions such as privacy, robustness, transparency, fairness, inclusiveness, continual learning, probabilistic models, approximation, computational complexity, and spectral complexity. We compare various vision transformer models based on their performance, the number of parameters, and the number of floating point operations (FLOPs) on multiple datasets.

While the Transformer architecture has become the de-facto standard for natural language processing tasks, its applications to computer vision remain limited. In vision, attention is either applied in conjunction with convolutional networks, or used to replace certain components of convolutional networks while keeping their overall structure in place. We show that this reliance on CNNs is not necessary and a pure transformer applied directly to sequences of image patches can perform very well on image classification tasks. When pre-trained on large amounts of data and transferred to multiple mid-sized or small image recognition benchmarks (ImageNet, CIFAR-100, VTAB, etc.), Vision Transformer (ViT) attains excellent results compared to state-of-the-art convolutional networks while requiring substantially fewer computational resources to train.

Self-supervised learning methods overcome the key bottleneck for building more capable AI: limited availability of labeled data. However, one of the drawbacks of self-supervised architectures is that the representations that they learn are implicit and it is hard to extract meaningful information about the encoded world states, such as 3D structure of the visual scene encoded in a depth map. Moreover, in the visual domain such representations only rarely undergo evaluations that may be critical for downstream tasks, such as vision for autonomous cars. Herein, we propose a framework for evaluating visual representations for illumination invariance in the context of depth perception. We develop a new predictive coding-based architecture and a hybrid fully-supervised/self-supervised learning method. We propose a novel architecture that extends the predictive coding approach: PRedictive Lateral bottom-Up and top-Down Encoder-decoder Network (PreludeNet), which explicitly learns to infer and predict depth from video frames. In PreludeNet, the encoder's stack of predictive coding layers is trained in a self-supervised manner, while the predictive decoder is trained in a supervised manner to infer or predict the depth. We evaluate the robustness of our model on a new synthetic dataset, in which lighting conditions (such as overall illumination, and effect of shadows) can be be parametrically adjusted while keeping all other aspects of the world constant. PreludeNet achieves both competitive depth inference performance and next frame prediction accuracy. We also show how this new network architecture, coupled with the hybrid fully-supervised/self-supervised learning method, achieves balance between the said performance and invariance to changes in lighting. The proposed framework for evaluating visual representations can be extended to diverse task domains and invariance tests.

In this paper, we propose a Vision-Audio-Language Omni-peRception pretraining model (VALOR) for multi-modal understanding and generation. Different from widely-studied vision-language pretraining models, VALOR jointly models relationships of vision, audio and language in an end-to-end manner. It contains three separate encoders for single modality representations, and a decoder for multimodal conditional text generation. We design two pretext tasks to pretrain VALOR model, including Multimodal Grouping Alignment (MGA) and Multimodal Grouping Captioning (MGC). MGA projects vision, language and audio to the same common space, building vision-language, audio-language and audiovisual-language alignment simultaneously. MGC learns how to generate text tokens in conditions of vision, audio or their both. To promote vision-audio-language pretraining research, we construct a large-scale high-quality tri-modality dataset named VALOR-1M, which contains 1M audiable videos with human annotated audiovisual captions. Extensive experiments show that VALOR can learn strong multimodal correlations and be generalized to various downstream tasks (e.g., retrieval, captioning and question answering), with different input modalities (e.g., vision-language, audio-language and audiovisual-language). VALOR achieves new state-of-the-art performances on series of public cross-modality benchmarks. Code and data are available at project page https://casia-iva-group.github.io/projects/VALOR.

Vision-Language Models (VLMs) leverage aligned visual encoders to transform images into visual tokens, allowing them to be processed similarly to text by the backbone large language model (LLM). This unified input paradigm enables VLMs to excel in vision-language tasks such as visual question answering (VQA). To improve fine-grained visual reasoning, recent advancements in vision-language modeling introduce image cropping techniques that feed all encoded sub-images into the model. However, this approach significantly increases the number of visual tokens, leading to inefficiency and potential distractions for the LLM. To address the generalization challenges of image representation in VLMs, we propose a lightweight, universal framework that seamlessly integrates with existing VLMs to enhance their ability to process finegrained details. Our method leverages textual semantics to identify key visual areas, improving VQA performance without requiring any retraining of the VLM. Additionally, it incorporates textual signals into the visual encoding process, enhancing both efficiency and effectiveness. The proposed method, SEMCLIP, strengthens the visual understanding of a 7B VLM, LLaVA-1.5 by 3.3% on average across 7 benchmarks, and particularly by 5.3% on the challenging detailed understanding benchmark V*.

Vision-and-language (VL) models with separate encoders for each modality (e.g., CLIP) have become the go-to models for zero-shot image classification and image-text retrieval. The bulk of the evaluation of these models is, however, performed with English text only: the costly creation of language-specific image-caption datasets has limited multilingual VL benchmarks to a handful of high-resource languages. In this work, we introduce Babel-ImageNet, a massively multilingual benchmark that offers (partial) translations of 1000 ImageNet labels to 92 languages, built without resorting to machine translation (MT) or requiring manual annotation. We instead automatically obtain reliable translations of ImageNext concepts by linking them -- via shared WordNet synsets -- to BabelNet, a massively multilingual lexico-semantic network. We evaluate 8 different publicly available multilingual CLIP models on zero-shot image classification (ZS-IC) for each of the 92 Babel-ImageNet languages, demonstrating a significant gap between English ImageNet performance and that of high-resource languages (e.g., German or Chinese), and an even bigger gap for low-resource languages (e.g., Sinhala or Lao). Crucially, we show that the models' ZS-IC performance on Babel-ImageNet highly correlates with their performance in image-text retrieval, validating that Babel-ImageNet is suitable for estimating the quality of the multilingual VL representation spaces for the vast majority of languages that lack gold image-text data. Finally, we show that the performance of multilingual CLIP for low-resource languages can be drastically improved via cheap, parameter-efficient language-specific training. We make our code and data publicly available: https://github.com/gregor-ge/Babel-ImageNet

Recent advancements in multimodal fusion have witnessed the remarkable success of vision-language (VL) models, which excel in various multimodal applications such as image captioning and visual question answering. However, building VL models requires substantial hardware resources, where efficiency is restricted by two key factors: the extended input sequence of the language model with vision features demands more computational operations, and a large number of additional learnable parameters increase memory complexity. These challenges significantly restrict the broader applicability of such models. To bridge this gap, we propose ADEM-VL, an efficient vision-language method that tunes VL models based on pretrained large language models (LLMs) by adopting a parameter-free cross-attention mechanism for similarity measurements in multimodal fusion. This approach only requires embedding vision features into the language space, significantly reducing the number of trainable parameters and accelerating both training and inference speeds. To enhance representation learning in fusion module, we introduce an efficient multiscale feature generation scheme that requires only a single forward pass through the vision encoder. Moreover, we propose an adaptive fusion scheme that dynamically discards less relevant visual information for each text token based on its attention score. This ensures that the fusion process prioritizes the most pertinent visual features. With experiments on various tasks including visual question answering, image captioning, and instruction-following, we demonstrate that our framework outperforms existing approaches. Specifically, our method surpasses existing methods by an average accuracy of 0.77% on ScienceQA dataset, with reduced training and inference latency, demonstrating the superiority of our framework. The code is available at https://github.com/Hao840/ADEM-VL.

Vision Transformers (ViTs) have revolutionized computer vision by leveraging self-attention to model long-range dependencies. However, ViTs face challenges such as high computational costs due to the quadratic scaling of self-attention and the requirement of a large amount of training data. To address these limitations, we propose the Efficient Convolutional Vision Transformer (ECViT), a hybrid architecture that effectively combines the strengths of CNNs and Transformers. ECViT introduces inductive biases such as locality and translation invariance, inherent to Convolutional Neural Networks (CNNs) into the Transformer framework by extracting patches from low-level features and enhancing the encoder with convolutional operations. Additionally, it incorporates local-attention and a pyramid structure to enable efficient multi-scale feature extraction and representation. Experimental results demonstrate that ECViT achieves an optimal balance between performance and efficiency, outperforming state-of-the-art models on various image classification tasks while maintaining low computational and storage requirements. ECViT offers an ideal solution for applications that prioritize high efficiency without compromising performance.

We apply pre-trained architectures, originally developed for the ImageNet Large Scale Visual Recognition Challenge, for periocular recognition. These architectures have demonstrated significant success in various computer vision tasks beyond the ones for which they were designed. This work builds on our previous study using off-the-shelf Convolutional Neural Network (CNN) and extends it to include the more recently proposed Vision Transformers (ViT). Despite being trained for generic object classification, middle-layer features from CNNs and ViTs are a suitable way to recognize individuals based on periocular images. We also demonstrate that CNNs and ViTs are highly complementary since their combination results in boosted accuracy. In addition, we show that a small portion of these pre-trained models can achieve good accuracy, resulting in thinner models with fewer parameters, suitable for resource-limited environments such as mobiles. This efficiency improves if traditional handcrafted features are added as well.

Convolutional architectures have proven extremely successful for vision tasks. Their hard inductive biases enable sample-efficient learning, but come at the cost of a potentially lower performance ceiling. Vision Transformers (ViTs) rely on more flexible self-attention layers, and have recently outperformed CNNs for image classification. However, they require costly pre-training on large external datasets or distillation from pre-trained convolutional networks. In this paper, we ask the following question: is it possible to combine the strengths of these two architectures while avoiding their respective limitations? To this end, we introduce gated positional self-attention (GPSA), a form of positional self-attention which can be equipped with a ``soft" convolutional inductive bias. We initialise the GPSA layers to mimic the locality of convolutional layers, then give each attention head the freedom to escape locality by adjusting a gating parameter regulating the attention paid to position versus content information. The resulting convolutional-like ViT architecture, ConViT, outperforms the DeiT on ImageNet, while offering a much improved sample efficiency. We further investigate the role of locality in learning by first quantifying how it is encouraged in vanilla self-attention layers, then analysing how it is escaped in GPSA layers. We conclude by presenting various ablations to better understand the success of the ConViT. Our code and models are released publicly at https://github.com/facebookresearch/convit.

Vision Transformers (ViTs) have achieved overwhelming success, yet they suffer from vulnerable resolution scalability, i.e., the performance drops drastically when presented with input resolutions that are unseen during training. We introduce, ResFormer, a framework that is built upon the seminal idea of multi-resolution training for improved performance on a wide spectrum of, mostly unseen, testing resolutions. In particular, ResFormer operates on replicated images of different resolutions and enforces a scale consistency loss to engage interactive information across different scales. More importantly, to alternate among varying resolutions effectively, especially novel ones in testing, we propose a global-local positional embedding strategy that changes smoothly conditioned on input sizes. We conduct extensive experiments for image classification on ImageNet. The results provide strong quantitative evidence that ResFormer has promising scaling abilities towards a wide range of resolutions. For instance, ResFormer-B-MR achieves a Top-1 accuracy of 75.86% and 81.72% when evaluated on relatively low and high resolutions respectively (i.e., 96 and 640), which are 48% and 7.49% better than DeiT-B. We also demonstrate, moreover, ResFormer is flexible and can be easily extended to semantic segmentation, object detection and video action recognition. Code is available at https://github.com/ruitian12/resformer.

Vision transformers have achieved remarkable success in computer vision tasks by using multi-head self-attention modules to capture long-range dependencies within images. However, the high inference computation cost poses a new challenge. Several methods have been proposed to address this problem, mainly by slimming patches. In the inference stage, these methods classify patches into two classes, one to keep and the other to discard in multiple layers. This approach results in additional computation at every layer where patches are discarded, which hinders inference acceleration. In this study, we tackle the patch slimming problem from a different perspective by proposing a life regression module that determines the lifespan of each image patch in one go. During inference, the patch is discarded once the current layer index exceeds its life. Our proposed method avoids additional computation and parameters in multiple layers to enhance inference speed while maintaining competitive performance. Additionally, our approach requires fewer training epochs than other patch slimming methods.

This paper studies a conceptually simple extension of Masked Autoencoders (MAE) to spatiotemporal representation learning from videos. We randomly mask out spacetime patches in videos and learn an autoencoder to reconstruct them in pixels. Interestingly, we show that our MAE method can learn strong representations with almost no inductive bias on spacetime (only except for patch and positional embeddings), and spacetime-agnostic random masking performs the best. We observe that the optimal masking ratio is as high as 90% (vs. 75% on images), supporting the hypothesis that this ratio is related to information redundancy of the data. A high masking ratio leads to a large speedup, e.g., > 4x in wall-clock time or even more. We report competitive results on several challenging video datasets using vanilla Vision Transformers. We observe that MAE can outperform supervised pre-training by large margins. We further report encouraging results of training on real-world, uncurated Instagram data. Our study suggests that the general framework of masked autoencoding (BERT, MAE, etc.) can be a unified methodology for representation learning with minimal domain knowledge.

Vision-language models have revolutionized human-computer interaction and shown significant progress in multi-modal tasks. However, applying these models to complex visual tasks like medical image analysis remains challenging. In this study, we propose ViLaM, a unified Vision-Language transformer model that integrates instruction tuning predicated on a large language model. This approach enables us to optimally utilize the knowledge and reasoning capacities of large pre-trained language models for an array of tasks encompassing both language and vision. We employ frozen pre-trained encoders to encode and align both image and text features, enabling ViLaM to handle a variety of visual tasks following textual instructions. Besides, we've designed cycle training for referring expressions to address the need for high-quality, paired referring expression datasets for training large models in terms of both quantity and quality. We evaluated ViLaM's exceptional performance on public general datasets and further confirmed its generalizability on medical datasets. Importantly, we've observed the model's impressive zero-shot learning ability, indicating the potential future application of ViLaM in the medical field.