Image Restoration Hub

A Image restoration hub based on basicsr library, including: image super resolution, image denoise, etc. All the settings are adjusted on our un-public infrared dataset.

News

• 2024-10-14. Add inference code to quickly verify model result.

Features

Supported models

• ECBSR. Real-time super resolution.
• SwinIR. A swin-tansformer based strong image resotration baseline.

Supported Metrics

• niqe
• ssim
• psnr

Supported Losses

• L1 loss
• MSE loss
• SSIM loss
• MSSSIM loss
• MultiScaleGAN loss

Supported Input Type

• grayscale
• rgb

Results

Subjective effect.

Enviroments

• python=3.10.14
• torch=2.4.0
• torchvision=0.19.0

Installation

pip install -r requirements.txt
python setup.py develop