KPN-Single-Image

A PyTorch implementation of kernel prediction network for single image denoising.

1 Samples

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

GT | Input | Denoised by KPN-Single-Image

2 Training

Trained models are available via this OneDrive link

If you want to train your own data, change arg baseroot to your own data path, then run:

sh run.sh

3 Validation

We only provide one kind of model for specific noise level. If you want to test your own data, change the arg baseroot to the path to your validation set, save_name to saving path, and load_name to trained model path.

python validation.py

4 Acknowledgement

This KPN code is borrowed from the project.

@inproceedings{mildenhall2018burst,
  title={Burst denoising with kernel prediction networks},
  author={Mildenhall, Ben and Barron, Jonathan T and Chen, Jiawen and Sharlet, Dillon and Ng, Ren and Carroll, Robert},
  booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition},
  pages={2502--2510},
  year={2018}
}