https://github.com/sczhou/ignn
[NeurIPS 2020] Cross-Scale Internal Graph Neural Network for Image Super-Resolution
https://github.com/sczhou/ignn
deep-learning image-preprocessing image-restoration super-resolution
Last synced: 6 months ago
JSON representation
[NeurIPS 2020] Cross-Scale Internal Graph Neural Network for Image Super-Resolution
- Host: GitHub
- URL: https://github.com/sczhou/ignn
- Owner: sczhou
- Created: 2020-06-29T16:57:33.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2021-02-23T12:58:33.000Z (over 4 years ago)
- Last Synced: 2025-04-09T21:17:48.790Z (6 months ago)
- Topics: deep-learning, image-preprocessing, image-restoration, super-resolution
- Language: Python
- Homepage:
- Size: 146 KB
- Stars: 313
- Watchers: 18
- Forks: 40
- Open Issues: 11
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# IGNN
Code repo for "Cross-Scale Internal Graph Neural Network for Image Super-Resolution" [[paper]](https://proceedings.neurips.cc/paper/2020/file/23ad3e314e2a2b43b4c720507cec0723-Paper.pdf) [[supp]](https://proceedings.neurips.cc/paper/2020/file/23ad3e314e2a2b43b4c720507cec0723-Supplemental.pdf)
## Prepare datasets
1 Download training dataset and test datasets from [here](https://drive.google.com/file/d/1fFBCXkUIgHkjqWiCeW7w-1TYHE0A2ZZF/view?usp=sharing).
2 Crop training dataset DIV2K to sub-images.
```
python ./datasets/prepare_DIV2K_subimages.py
```
Remember to modify the 'input_folder' and 'save_folder' in the above script.
## Dependencies and Installation
The denoising code is tested with Python 3.7, PyTorch 1.1.0 and Cuda 9.0 but is likely to run with newer versions of PyTorch and Cuda.
1 Create conda environment.
```
conda create --name ignn
conda activate ignn
conda install pytorch=1.1.0 torchvision=0.3.0 cudatoolkit=9.0 -c pytorch
```
2 Install PyInn.
```
pip install git+https://github.com/szagoruyko/pyinn.git@master
```
3 Install matmul_cuda.
```
bash install.sh
```
4 Install other dependencies.
```
pip install -r requirements.txt
```
## Pretrained Models
Downloading the pretrained models from this [link](https://drive.google.com/drive/folders/1xS0jATn0MddZkLl2Rx9VPLh-U_rUxjt1?usp=sharing) and put them into ./ckpt
## Training
Use the following command to train the network:
```
python runner.py
--gpu [gpu_id]\
--phase 'train'\
--scale [2/3/4]\
--dataroot [dataset root]\
--out [output path]
```
Use the following command to resume training the network:
```
python runner.py
--gpu [gpu_id]\
--phase 'resume'\
--weights './ckpt/IGNN_x[2/3/4].pth'\
--scale [2/3/4]\
--dataroot [dataset root]\
--out [output path]
```
You can also use the following simple command with different settings in config.py:
```
python runner.py
```
## Testing
Use the following command to test the network on benchmark datasets (w/ GT):
```
python runner.py \
--gpu [gpu_id]\
--phase 'test'\
--weights './ckpt/IGNN_x[2/3/4].pth'\
--scale [2/3/4]\
--dataroot [dataset root]\
--testname [Set5, Set14, BSD100, Urban100, Manga109]\
--out [output path]
```
Use the following command to test the network on your demo images (w/o GT):
```
python runner.py \
--gpu [gpu_id]\
--phase 'test'\
--weights './ckpt/IGNN_x[2/3/4].pth'\
--scale [2/3/4]\
--demopath [test folder path]\
--testname 'Demo'\
--out [output path]
```
You can also use the following simple command with different settings in config.py:
```
python runner.py
```
## Visual Results (x4)
For visual comparison on the 5 benchmarks, you can download our IGNN results from [here](https://drive.google.com/file/d/15x81tYQVpml4OvFqbA05mQQSRKL8phxz/view?usp=sharing).
### Some examples
## Citation
If you find our work useful for your research, please consider citing the following papers :)
```
@inproceedings{zhou2020cross,
title={Cross-scale internal graph neural network for image super-resolution},
author={Zhou, Shangchen and Zhang, Jiawei and Zuo, Wangmeng and Loy, Chen Change},
booktitle={Advances in Neural Information Processing Systems},
year={2020}
}
```
## Contact
We are glad to hear from you. If you have any questions, please feel free to contact shangchenzhou@gmail.com.
## License
This project is open sourced under MIT license.