| # LayoutLM | |
| **Multimodal (text + layout/format + image) pre-training for document AI** | |
| - April 17th, 2021: [LayoutXLM](https://arxiv.org/abs/2104.08836) extends the LayoutLM/LayoutLMv2 into multilingual support! In addition, we also introduce XFUN, a multilingual form understanding benchmark including forms with human labeled key-value pairs in 7 languages (Chinese, Japanese, Spanish, French, Italian, German, Portuguese). | |
| - December 29th, 2020: [LayoutLMv2](https://arxiv.org/abs/2012.14740) is coming with the new SOTA on a wide varierty of document AI tasks, including [DocVQA](https://rrc.cvc.uab.es/?ch=17&com=evaluation&task=1) and [SROIE](https://rrc.cvc.uab.es/?ch=13&com=evaluation&task=3) leaderboard. | |
| ## Introduction | |
| LayoutLM is a simple but effective multi-modal pre-training method of text, layout and image for visually-rich document understanding and information extraction tasks, such as form understanding and receipt understanding. LayoutLM archives the SOTA results on multiple datasets. For more details, please refer to our paper: | |
| [LayoutLM: Pre-training of Text and Layout for Document Image Understanding](https://arxiv.org/abs/1912.13318) | |
| Yiheng Xu, Minghao Li, Lei Cui, Shaohan Huang, Furu Wei, Ming Zhou, [KDD 2020](https://www.kdd.org/kdd2020/accepted-papers) | |
| [LayoutLMv2: Multi-modal Pre-training for Visually-Rich Document Understanding](https://arxiv.org/abs/2012.14740) | |
| Yang Xu, Yiheng Xu, Tengchao Lv, Lei Cui, Furu Wei, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Wanxiang Che, Min Zhang, Lidong Zhou, [ACL 2021](#) | |
| [LayoutXLM: Multimodal Pre-training for Multilingual Visually-rich Document Understanding](https://arxiv.org/abs/2104.08836) | |
| Yiheng Xu, Tengchao Lv, Lei Cui, Guoxin Wang, Yijuan Lu, Dinei Florencio, Cha Zhang, Furu Wei, [Preprint](#) | |
| ## Release Notes | |
| **\*\*\*\*\* New Aug 7th, 2020: Our new document understanding datasets, [TableBank](https://doc-analysis.github.io/tablebank-page/) (LREC 2020) and [DocBank](https://doc-analysis.github.io/docbank-page/) (COLING 2020), are now publicly available.\*\*\*\*\*** | |
| **\*\*\*\*\* New May 16th, 2020: Our LayoutLM paper has been accepted to KDD 2020 as a full paper in the research track\*\*\*\*\*** | |
| **\*\*\*\*\* New Feb 18th, 2020: Initial release of pre-trained models and fine-tuning code for LayoutLM v1 \*\*\*\*\*** | |
| ## Pre-trained Model | |
| We pre-train LayoutLM on IIT-CDIP Test Collection 1.0\* dataset with two settings. | |
| * LayoutLM-Base, Uncased (11M documents, 2 epochs): 12-layer, 768-hidden, 12-heads, 113M parameters || [OneDrive](https://1drv.ms/u/s!ApPZx_TWwibInS3JD3sZlPpQVZ2b?e=bbTfmM) | [Google Drive](https://drive.google.com/open?id=1Htp3vq8y2VRoTAwpHbwKM0lzZ2ByB8xM) | |
| * LayoutLM-Large, Uncased (11M documents, 2 epochs): 24-layer, 1024-hidden, 16-heads, 343M parameters || [OneDrive](https://1drv.ms/u/s!ApPZx_TWwibInSy2nj7YabBsTWNa?e=p4LQo1) | [Google Drive](https://drive.google.com/open?id=1tatUuWVuNUxsP02smZCbB5NspyGo7g2g) | |
| \*As some downstream datasets are the subsets of IIT-CDIP, we have carefully excluded the overlap portion from the pre-training data. | |
| ## Fine-tuning Example | |
| We evaluate LayoutLM on several document image understanding datasets, and it outperforms several SOTA pre-trained models and approaches. | |
| Setup environment as follows: | |
| ~~~bash | |
| conda create -n layoutlm python=3.6 | |
| conda activate layoutlm | |
| conda install pytorch==1.4.0 cudatoolkit=10.1 -c pytorch | |
| git clone https://github.com/NVIDIA/apex && cd apex | |
| pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" ./ | |
| pip install . | |
| ## For development mode | |
| # pip install -e ".[dev]" | |
| ~~~ | |
| ### Sequence Labeling Task | |
| We give a fine-tuning example for sequence labeling tasks. You can run this example on [FUNSD](https://guillaumejaume.github.io/FUNSD/), a dataset for document understanding tasks. | |
| First, we need to preprocess the JSON file into txt. You can run the preprocessing scripts `funsd_preprocess.py` in the `scripts` directory. For more options, please refer to the arguments. | |
| ~~~bash | |
| cd examples/seq_labeling | |
| ./preprocess.sh | |
| ~~~ | |
| After preprocessing, run LayoutLM as follows: | |
| ~~~bash | |
| python run_seq_labeling.py --data_dir data \ | |
| --model_type layoutlm \ | |
| --model_name_or_path path/to/pretrained/model/directory \ | |
| --do_lower_case \ | |
| --max_seq_length 512 \ | |
| --do_train \ | |
| --num_train_epochs 100.0 \ | |
| --logging_steps 10 \ | |
| --save_steps -1 \ | |
| --output_dir path/to/output/directory \ | |
| --labels data/labels.txt \ | |
| --per_gpu_train_batch_size 16 \ | |
| --per_gpu_eval_batch_size 16 \ | |
| --fp16 | |
| ~~~ | |
| Note: The `DataParallel` will be enabled automatically to utilize all GPUs. If you want to train with `DistributedDataParallel`, please run the script like: | |
| ~~~bash | |
| # Suppose you have 4 GPUs. | |
| python -m torch.distributed.launch --nproc_per_node=4 run_seq_labeling.py --data_dir data \ | |
| --model_type layoutlm \ | |
| --model_name_or_path path/to/pretrained/model/directory \ | |
| --do_lower_case \ | |
| --max_seq_length 512 \ | |
| --do_train \ | |
| --num_train_epochs 100.0 \ | |
| --logging_steps 10 \ | |
| --save_steps -1 \ | |
| --output_dir path/to/output/directory \ | |
| --labels data/labels.txt \ | |
| --per_gpu_train_batch_size 16 \ | |
| --per_gpu_eval_batch_size 16 \ | |
| --fp16 | |
| ~~~ | |
| Then you can do evaluation or inference by replacing `--do_train` with `--do_eval` or `--do_predict` | |
| Also, you can run Bert and RoBERTa baseline by modifying the `--model_type` argument. For more options, please refer to the arguments of `run.py`. | |
| ### Document Image Classification Task | |
| We also fine-tune LayoutLM on the document image classification task. You can download the [RVL-CDIP](https://www.cs.cmu.edu/~aharley/rvl-cdip/) dataset from [here](https://www.cs.cmu.edu/~aharley/rvl-cdip/). Because this dataset only provides the document image, you should use the OCR tool to get the texts and bounding boxes. For example, you can easily use Tesseract, an open-source OCR engine, to generate corresponding OCR data in hOCR format. For more details, please refer to the [Tesseract wiki](https://github.com/tesseract-ocr/tesseract/wiki). Your processed data should look like [this sample data](https://1drv.ms/u/s!ApPZx_TWwibInTlBa5q3tQ7QUdH_?e=UZLVFw). | |
| With the processed OCR data, you can run LayoutLM as follows: | |
| ~~~bash | |
| python run_classification.py --data_dir data \ | |
| --model_type layoutlm \ | |
| --model_name_or_path path/to/pretrained/model/directory \ | |
| --output_dir path/to/output/directory \ | |
| --do_lower_case \ | |
| --max_seq_length 512 \ | |
| --do_train \ | |
| --do_eval \ | |
| --num_train_epochs 40.0 \ | |
| --logging_steps 5000 \ | |
| --save_steps 5000 \ | |
| --per_gpu_train_batch_size 16 \ | |
| --per_gpu_eval_batch_size 16 \ | |
| --evaluate_during_training \ | |
| --fp16 | |
| ~~~ | |
| Similarly, you can do evaluation by changing `--do_train` to `--do_eval` and `--do_test` | |
| Like the sequence labeling task, you can run Bert and RoBERTa baseline by modifying the `--model_type` argument. | |
| ### Results | |
| #### SROIE (field-level) | |
| | Model | Hmean | | |
| | ------------------------------------------------------------ | ---------- | | |
| | BERT-Large | 90.99% | | |
| | RoBERTa-Large | 92.80% | | |
| | [Ranking 1st in SROIE](https://rrc.cvc.uab.es/?ch=13&com=evaluation&task=3) | 94.02% | | |
| | [**LayoutLM**](https://rrc.cvc.uab.es/?ch=13&com=evaluation&view=method_info&task=3&m=71448) | **96.04%** | | |
| #### RVL-CDIP | |
| | Model | Accuracy | | |
| | ------------------------------------------------------------ | ---------- | | |
| | BERT-Large | 89.92% | | |
| | RoBERTa-Large | 90.11% | | |
| | [VGG-16 (Afzal et al., 2017)](https://arxiv.org/abs/1704.03557) | 90.97% | | |
| | [Stacked CNN Ensemble (Das et al., 2018)](https://arxiv.org/abs/1801.09321) | 92.21% | | |
| | [LadderNet (Sarkhel & Nandi, 2019)](https://www.ijcai.org/Proceedings/2019/0466.pdf) | 92.77% | | |
| | [Multimodal Ensemble (Dauphinee et al., 2019)](https://arxiv.org/abs/1912.04376) | 93.07% | | |
| | **LayoutLM** | **94.42%** | | |
| #### FUNSD (field-level) | |
| | Model | Precision | Recall | F1 | | |
| | ------------- | ---------- | ---------- | ---------- | | |
| | BERT-Large | 0.6113 | 0.7085 | 0.6563 | | |
| | RoBERTa-Large | 0.6780 | 0.7391 | 0.7072 | | |
| | **LayoutLM** | **0.7677** | **0.8195** | **0.7927** | | |
| ## Citation | |
| If you find LayoutLM useful in your research, please cite the following paper: | |
| ``` latex | |
| @misc{xu2019layoutlm, | |
| title={LayoutLM: Pre-training of Text and Layout for Document Image Understanding}, | |
| author={Yiheng Xu and Minghao Li and Lei Cui and Shaohan Huang and Furu Wei and Ming Zhou}, | |
| year={2019}, | |
| eprint={1912.13318}, | |
| archivePrefix={arXiv}, | |
| primaryClass={cs.CL} | |
| } | |
| ``` | |
| ## License | |
| This project is licensed under the license found in the LICENSE file in the root directory of this source tree. | |
| Portions of the source code are based on the [transformers](https://github.com/huggingface/transformers) project. | |
| [Microsoft Open Source Code of Conduct](https://opensource.microsoft.com/codeofconduct) | |
| ### Contact Information | |
| For help or issues using LayoutLM, please submit a GitHub issue. | |
| For other communications related to LayoutLM, please contact Lei Cui (`[email protected]`), Furu Wei (`[email protected]`). | |