PyTorch Implementation of MPRNet.
Deep learning model to remove rain from the images.
This model achieves 26 dB PSNR on Rain100H Dataset.
Input 👉 restored image 👉 ground-truth
More examples are in the Results directory.
- This model is trained on Rain Dataset.
- The model is optimized using two losses:
Charbonnier Loss and Laplacian Edge Loss. - All the training images were resized to
64x64to fit them into a single GPU. - Learning rate is set to
2e-4and it is decayed overtime to1e-6. - Trained model is available in
Pretrained modeldirectory.