Add metadata and improve model card

This PR adds the `image-feature-extraction` pipeline tag to the model card metadata and ensures the README is properly formatted. This will improve the visibility and discoverability of the model on the Hugging Face Hub.

README.md

@@ -1,24 +1,20 @@
+---
+pipeline_tag: image-feature-extraction
+---
+
 <div align="center">
 <h1>WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens</h1>
 
-[![arXiv](https://img.shields.io/badge/arXiv-2508.05599-b31b1b.svg)](https://arxiv.org/abs/2605.18115)
-[![Github](https://img.shields.io/badge/Github-WeTok-blue)](https://github.com/markywg/WinTok)
+[![arXiv](https://img.shields.io/badge/arXiv-2605.18115-b31b1b.svg)](https://arxiv.org/abs/2605.18115)
+[![Github](https://img.shields.io/badge/Github-WinTok-blue)](https://github.com/markywg/WinTok)
 [![Hugging Face Model](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-yellow)](https://huggingface.co/markyw/WinTok/tree/main)
 </div>
 
 This project introduces **WinTok**, a concise hybrid visual tokenizer designed to resolve the long-standing conflict between visual understanding and generation. By decoupling semantic and pixel tokens with an asymmetric distillation mechanism, WinTok achieves a win-win across reconstruction, understanding, and generation, surpassing strong baselines with substantially less training data. <br><br>
 
-> <a href="https://arxiv.org/abs/2605.18115">WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens</a><br>
-> [Yiwei Guo](https://scholar.google.com/citations?user=HCAyeJIAAAAJ&hl=zh-CN&oi=ao), [Shaobin Zhuang](https://scholar.google.com/citations?user=PGaDirMAAAAJ&hl=zh-CN&oi=ao), Canmiao Fu, [Zhipeng Huang](https://scholar.google.com/citations?user=_fnuIHUAAAAJ&hl=zh-CN&oi=ao), [Chen Li](https://scholar.google.com/citations?hl=zh-CN&user=WDJL3gYAAAAJ), Jing LYU, [Yali Wang](https://scholar.google.com/citations?hl=zh-CN&user=hD948dkAAAAJ)<br>
+> [WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens](https://huggingface.co/papers/2605.18115)<br>
+> Yiwei Guo, Shaobin Zhuang, Canmiao Fu, Zhipeng Huang, Chen Li, Jing LYU, Yali Wang<br>
 > Shenzhen Institutes of Advanced Technology (Chinese Academy of Sciences), WeChat Vision (Tencent Inc.), Shanghai Jiao Tong University<br>
-> ```
-> @article{guo2026wintok,
->   title={WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens},
->   author={Guo, Yiwei and Zhuang, Shaobin and Huang, Zhipeng and Fu, Canmiao and Li, Chen and LYU, Jing and Wang, Yali},
->   journal={arXiv preprint arXiv:2605.18115},
->   year={2026}
-> }
-> ```
 
 <p align="center">
   <img src="./assets/visualization.jpg" width="90%">
@@ -33,7 +29,7 @@ This project introduces **WinTok**, a concise hybrid visual tokenizer designed t
 
 ### 🛠️ Installation
 - **Dependencies**: 
-```
+```bash
 bash env.sh
 ```
 
@@ -49,7 +45,7 @@ imagenet
 ```
 
 Run the 256×256 resolution evaluation script, change the corresponding path:
-```
+```bash
 bash scripts/eval_tokenizer/eval_metrics_ddp.sh
 ```
 
@@ -63,7 +59,7 @@ MSCOCO2017
 ```
 
 Run the 256×256 resolution evaluation script, change the corresponding path:
-```
+```bash
 bash scripts/eval_tokenizer/eval_metrics_ddp.sh
 ```
 
@@ -71,6 +67,17 @@ bash scripts/eval_tokenizer/eval_metrics_ddp.sh
 ### Inference
 
 Simply test the effect of model reconstruction:
-```
+```bash
 python recon.py --ckpt_path path_to_ckpt
+```
+
+## Citation
+
+```bibtex
+@article{guo2026wintok,
+  title={WinTok: A Win-Win Hybrid Tokenizer via Decomposing Visual Understanding and Generation with Transferable Tokens},
+  author={Guo, Yiwei and Zhuang, Shaobin and Huang, Zhipeng and Fu, Canmiao and Li, Chen and LYU, Jing and Wang, Yali},
+  journal={arXiv preprint arXiv:2605.18115},
+  year={2026}
+}
 ```