Keras implementation of the method:
LSD2 -- Joint Denoising and Deblurring of Short and Long Exposure Images with CNNs [arXiv]

Download the weights for LSD2 from here and put them to the checkpoints folder.

To process the images in the folder input/test/..., execute:
python predict.py

Output images will be saved to output/test/.... Note that the network expects a pair of short and long exposure images taken with specific camera settings. It will probably not work with other kinds of inputs.

To train your own model, execute:
python train.py

The folders input/train/... and input/val/... contain a few training and validation images. You can replace them with your own.

Soon to be uploaded.

If you find our code helpful in your research or work, please cite our paper.

@InProceedings{Mustaniemi_2020_BMVC,
  author = {Mustaniemi, Janne and Kannala, Juho and Matas, Jiri and Särkkä, Simo and Heikkilä, Janne},
  title = {LSD_2 - Joint Denoising and Deblurring of Short and Long Exposure Images with CNNs},
  booktitle = {The 31st British Machine Vision Virtual Conference (BMVC)},
  month = {September},
  year = {2020}
}