Add metadata for license, library_name, and pipeline_tag

README.md

@@ -1,3 +1,14 @@
+---
+license: apache-2.0
+library_name: diffusers
+pipeline_tag: image-to-image
+tags:
+- super-resolution
+- image-restoration
+- dpo
+- one-step-generation
+---
+
 <div align="center">
 <h2>GDPO-SR: Group Direct Preference Optimization for One-Step Generative Image Super-Resolution</h2>
 
@@ -11,8 +22,9 @@
 <sup>1</sup>The Hong Kong Polytechnic University, <sup>2</sup>OPPO Research Institute
 </div>
 
-[![](https://img.shields.io/badge/ArXiv%20-Paper-b31b1b?logo=arxiv&logoColor=red)](https://arxiv.org/pdf/2603.16769)&nbsp; [![weights](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-model%20weights-blue)](https://huggingface.co/Joypop/GDPO/tree/main)
+[![](https://img.shields.io/badge/ArXiv%20-Paper-b31b1b?logo=arxiv&logoColor=red)](https://huggingface.co/papers/2603.16769)&nbsp; [![weights](https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-model%20weights-blue)](https://huggingface.co/Joypop/GDPO/tree/main)
 
+This repository contains the weights for GDPO-SR, presented in the paper [GDPO-SR: Group Direct Preference Optimization for One-Step Generative Image Super-Resolution](https://huggingface.co/papers/2603.16769).
 
 ## ⏰ Update
 - **2026.3.19**: Paper is released on [ArXiv](https://arxiv.org/pdf/2603.16769).
@@ -41,7 +53,7 @@ pip install -r requirements.txt
 
 #### Step 2: Prepare testing data and run testing command 
 You can modify input_path and output_path to run testing command. The input_path is the path of the test image and the output_path is the path where the output images are saved.
-```
+```shell
 CUDA_VISIBLE_DEVICES=0, python GDPOSR/inferences/test.py \
 --input_path test_LR \
 --output_path experiment/GDPOSR \
@@ -55,31 +67,31 @@ CUDA_VISIBLE_DEVICES=0, python GDPOSR/inferences/test.py \
 --time_step_noise=250 
 ```
 or 
-```
+```shell
 bash scripts/test/test.sh
 ```
 
 ## 🚄 Training Phase
 
 ### Step1: Prepare training data
-Download the [OpenImage dataset](https://storage.googleapis.com/openimages/web/index.html) and [LSIDR dataset](https://github.com/ofsoundof/LSDIR). For each image in the LSDIR dataset, crop multiple 512×512 image patches using a sliding window with a stride of 64 pixels;
+Download the [LSIDR dataset](https://github.com/ofsoundof/LSDIR) and [FFHQ dataset](https://github.com/NVlabs/ffhq-dataset) and crop multiple 512×512 image patches using a sliding window with a stride of 64 pixels;
 
 
 ### Step2: Train NAOSD.
-```
+```shell
 bash scripts/train/train_NAOSD.sh
 ```
-The hyperparameters in train_NAOSD.sh can be modified to suit different experimental settings. Besides, after training with NAOSD, you can use GDPOSR/mergelora.py to merge the LoRA into the UNet and VAE as base model for subsequent reinforcement learning training and inference.
+The hyperparameters in train_NAOSD.sh can be modified to suit different experimental settings. Besides, after training with NAOSD, you can use `GDPOSR/mergelora.py` to merge the LoRA into the UNet and VAE as base model for subsequent reinforcement learning training and inference.
 
 ### Step3: Train GDPO-SR
-```
+```shell
 bash scripts/train/train_GDPOSR.sh
 ```
-The hyperparameters in train_GDPOSR.sh can be modified to suit different experimental settings. Besides, after training with GDPO-SR, you can use GDPOSR/mergelora.py to merge the LoRA into the UNet for subsequent inference.
+The hyperparameters in train_GDPOSR.sh can be modified to suit different experimental settings. Besides, after training with GDPO-SR, you can use `GDPOSR/mergelora.py` to merge the LoRA into the UNet for subsequent inference.
 
 ## 🔗 Citations
 
-```
+```bibtex
 @article{yi2026gdpo,
   title={GDPO-SR: Group Direct Preference Optimization for One-Step Generative Image Super-Resolution},
   author={Yi, Qiaosi and Li, Shuai and Wu, Rongyuan and Sun, Lingchen and Zhang, Zhengqiang and Zhang, Lei},
@@ -92,5 +104,4 @@ The hyperparameters in train_GDPOSR.sh can be modified to suit different experim
 This project is released under the [Apache 2.0 license](LICENSE).
 
 ## 📧 Contact
-If you have any questions, please contact: qiaosiyijoyies@gmail.com
-
+If you have any questions, please contact: qiaosiyijoyies@gmail.com