Improve model card with metadata, links, and structure
#1
by
nielsr
HF Staff
- opened
README.md
CHANGED
|
@@ -1,3 +1,55 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
pipeline_tag: image-to-image
|
| 4 |
+
library_name: diffusers
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers
|
| 8 |
+
|
| 9 |
+
<p align="center">
|
| 10 |
+
<a href="https://scholar.google.com.au/citations?user=GQzvqS4AAAAJ" target="_blank">Xingjian Leng</a><sup>1*</sup>   <b>·</b>  
|
| 11 |
+
<a href="https://1jsingh.github.io/" target="_blank">Jaskirat Singh</a><sup>1*</sup>   <b>·</b>  
|
| 12 |
+
<a href="https://hou-yz.github.io/" target="_blank">Yunzhong Hou</a><sup>1</sup>   <b>·</b>  
|
| 13 |
+
<a href="https://people.csiro.au/X/Z/Zhenchang-Xing/" target="_blank">Zhenchang Xing</a><sup>2</sup>  <b>·</b>  
|
| 14 |
+
<a href="https://www.sainingxie.com/" target="_blank">Saining Xie</a><sup>3</sup>  <b>·</b>  
|
| 15 |
+
<a href="https://zheng-lab-anu.github.io/" target="_blank">Liang Zheng</a><sup>1</sup> 
|
| 16 |
+
</p>
|
| 17 |
+
|
| 18 |
+
<p align="center">
|
| 19 |
+
<sup>1</sup> Australian National University   <sup>2</sup>Data61-CSIRO   <sup>3</sup>New York University   <br>
|
| 20 |
+
<sub><sup>*</sup>Project Leads  </sub>
|
| 21 |
+
</p>
|
| 22 |
+
|
| 23 |
+
[REPA-E: Unlocking VAE for End-to-End Tuning of Latent Diffusion Transformers](https://arxiv.org/abs/2504.10483)
|
| 24 |
+
|
| 25 |
+
Project Page: [https://end2end-diffusion.github.io](https://end2end-diffusion.github.io)
|
| 26 |
+
|
| 27 |
+
Code: [https://github.com/REPA-E/REPA-E](https://github.com/REPA-E/REPA-E)
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
## Model Description
|
| 31 |
+
|
| 32 |
+
We address the question of whether latent diffusion models and their VAE tokenizer can be trained end-to-end. REPA-E uses a representation-alignment (REPA) loss to enable stable and effective joint training of both components. This leads to significant training speedups (17x compared to REPA and 45x over vanilla training). End-to-end tuning also improves the VAE itself, resulting in a better latent structure and a drop-in replacement for existing VAEs (e.g., SD-VAE). Our method achieves state-of-the-art FID scores on ImageNet 256×256: 1.26 with CFG and 1.83 without CFG.
|
| 33 |
+
|
| 34 |
+
## Usage
|
| 35 |
+
|
| 36 |
+
See the GitHub repository for detailed instructions on environment setup, training, and evaluation. Pre-trained checkpoints are available on Hugging Face.
|
| 37 |
+
|
| 38 |
+
## Example Results
|
| 39 |
+
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
## Limitations and Bias
|
| 43 |
+
|
| 44 |
+
As with all diffusion models, REPA-E may exhibit biases present in the training data.
|
| 45 |
+
|
| 46 |
+
## Citation
|
| 47 |
+
|
| 48 |
+
```bibtex
|
| 49 |
+
@article{leng2025repae,
|
| 50 |
+
title={REPA-E: Unlocking VAE for End-to-End Tuning with Latent Diffusion Transformers},
|
| 51 |
+
author={Xingjian Leng and Jaskirat Singh and Yunzhong Hou and Zhenchang Xing and Saining Xie and Liang Zheng},
|
| 52 |
+
year={2025},
|
| 53 |
+
journal={arXiv preprint arXiv:2504.10483},
|
| 54 |
+
}
|
| 55 |
+
```
|