---
license: apache-2.0
---
# [CVPR2024 Oral] Image Processing GNN: Breaking Rigidity in Super-Resolution

<p align="left">
<a href="https://openaccess.thecvf.com/content/CVPR2024/papers/Tian_Image_Processing_GNN_Breaking_Rigidity_in_Super-Resolution_CVPR_2024_paper.pdf" alt="arXiv">
    Paper</a> | 
<a href="https://ipgnn.github.io" alt="arXiv">
    Project Page</a> |
<a href="https://openaccess.thecvf.com/content/CVPR2024/html/Tian_Image_Processing_GNN_Breaking_Rigidity_in_Super-Resolution_CVPR_2024_paper.html" alt="arXiv">
    <img src="https://img.shields.io/badge/CVPR'24-Oral-b31b31b.svg?style=flat" /></a>
<a href="https://huggingface.co/yuchuantian/IPG/tree/main" alt="Hugging Face Models">
    <img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Models-blue" /></a>
</p>
This is the official repo of our CVPR'24 paper **Image Processing GNN: Breaking Rigidity in Super-Resolution**. In the paper, we propose IPG: a Graph-based SR model that achieves outstanding performance on various SR benchmarks.

## News

7/6/2024: We opensourced the code & weights of IPG!

6/19/2024: Our work got the **Best Student Runner-up Award** of CVPR'24!🎉🎉

6/2/2024: We open-sourced [U-DiT](https://github.com/YuchuanTian/U-DiT), an efficient U-Net-style DiT variant.

## Replication

| Model | Scale | Urban100 | Weights                                                      |
| ----- | ----- | -------- | ------------------------------------------------------------ |
| GRL-B | 2x    | 33.97    | [🤗Link](https://huggingface.co/yuchuantian/IPG_rep/tree/main/GRLB_2_shbj1_676) |
| GRL-B | 3x    | 29.83    | [🤗Link](https://huggingface.co/yuchuantian/IPG_rep/tree/main/GRLB_3_shbj4_81) |
| GRL-B | 4x    | 27.53    | [🤗Link](https://huggingface.co/yuchuantian/IPG_rep/tree/main/GRLB_4_shbj4_68) |
| HAT   | 2x    | 34.45    | [🤗Link](https://huggingface.co/yuchuantian/IPG_rep/tree/main/HAT_2_shbj8_21) |