Pytorch implementation of "Blind Image Super-Resolution with Degradation-Aware Adaptation" (ACCV 2022).
- Python 3.6
- PyTorch >= 1.6.0
- numpy
- skimage
- imageio
- matplotlib
- cv2
1.1 Download the DIV2K dataset and the Flickr2K dataset.
1.2 Combine the HR images from these two datasets in your_data_path/DIV_FLI/HR
to build the DF2K dataset.
Run ./main_IMBD_nn_tt.py
to train on the DF2K dataset. Please update dir_data
in the option_IMBD_nn_tt.py
file as your_data_path
.
Download benchmark datasets (Set14) and prepare HR/LR images in your_data_path/benchmark
.
(Here, we recommand to generate the LR images with various blur kernel and noise from HR images in advance, and save the LR images in certain folders.)
Download RealSR(V3) and directly use the HR/LR images in your_data_path/RealSR(V3)/train_x4
.
Run ./Eval_IMBD_nn_tt.py
to test on benchmark dataset (one in-domain data and one out-domain data as example) and real world data with our pretrained model.
If you want to test on your model, please set the pre_train
as your_model_pth
in option_eval_IMBD_nn_tt.py
to load your model.
@InProceedings{Wang2022DAA,
author = {Wang, Yue and Ming, Jiawen and Jia, Xu and Elder, James and Lu, Huchuan},
title = {Blind Image Super-Resolution with Degradation-Aware Adaptation},
booktitle = {Proceedings of the asian conference on computer vision},
year = {2022},
}
This code is built on Unsupervised Degradation Representation Learning for Blind Super-Resolution (PyTorch). We thank the authors for sharing the codes.