IR_Hub

This repository is Hub of Image Restoration models. Refer the instruction below, you can implement sample process of training and testing.

※ 한국어 가이드는 README_kr.md 파일을 참조하시기 바랍니다.

Environment

• You must install a python virtual environment of version >= 3.8

conda create -n <env name> python=3.8

• I recommend that you install the PyTorch 1.12.1 + CUDA 11.6 version by following the instructions below.
• (If your model require another version of pytorch, you can choose suitable version.)

pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu116

• Other libraries can be installed with the following command.

pip install -r requirements.txt

Dataset

• Place your training dataset in the './trainsets/' path.

├── trainsets/
│    └── SR/
│         ├── HR/
│         │    └── DIV2K_train_HR/
│         │        ├── 0001.png
│         │        ├── ...
│         │            
│         └── LR/            
│              └── DIV2K_train_LR_bicubic/       
│                   ├── x2/  
│                   │     ├── 0001.png
│                   │     ├── ... 
│                   │ 
│                   ├── x3/
│                   └── x4/
│

• Place your testing dataset in the './testsets/' path.

├── testsets/
│    └── SR/
│         ├── HR/
│         │    ├── B100/
│         │    │   ├── 3096.png
│         │    │   ├── ...
│         │    │
│         │    └── Set5/ 
│         │
│         └── LR/            
│              ├── B100/       
│              │    ├── x2/  
│              │    │     ├── 3096.png
│              │    │     ├── ... 
│              │    │ 
│              │    ├── x3/
│              │    └── x4/
│              │
│              └── Set5/
│  

Training

• By running 'TRAIN_psnr.py', You can train image restoration model to improve PSNR.

python TRAIN_psnr.py --opt <Path of option file about model> 

• By following example commands, You can train the EDSR newtork.

python TRAIN_psnr.py --opt options/train_edsr_sr_baseline.json

Testing

• By running 'TEST.py', You can test image restoration model. Result of testing, You can get the PSNR, SSIM score and inference time.

python TEST.py --model <name of model> --scale <resolution scalig factor> --model_path <path of pretrained model> --folder_lq <path of low quality images> --folder_gt <path of grount truth images>

• By following example commands, You can test the EDSR newtork for x2 image super resolution.

python TEST.py --model edsr --scale 2 --model_path model_zoo/edsr_b2_x2.pth --folder_lq testsets/SR/LR/Set5/x2/ --folder_gt testsets/SR/HR/Set5/

Performance

• By following commands, You can check objective quality(PSNR,SSIM) and subjective quality(LPIPS) between generated iamges and ground truth images.

python EVAL_performance.py --folder_lq <path of generated image> --folder_gt <path of ground truth iamge>

Acknowledgement

• KAIR
• LPIPS