This code attempts to restore the degraded image mixed with Gaussian noise, Gaussian blur and JPEG compression from the perspective of hierarchical feature fusion.
The mixed degradation datasets is made in the same way as RL restore. In order to speed up the data reading, we save the data in npz format.
The code has been tested on CPU and GPU environments.
- Python 3.7
- PyTorch 1.5
python train.py -g 1test on a standard test set
python test.py -g 1test on your datasets
python test_for_your_data.py -g 1