CVF-SID_PyTorch

This repository contains the official code to reproduce the results from the paper:

CVF-SID: Cyclic multi-Variate Function for Self-Supervised Image Denoising by Disentangling Noise from Image

[arXiv] [presentation]

Installation

Clone this repository into any place you want.

git clone https://github.com/Reyhanehne/CVF-SID_PyTorch.git
cd CVF-SID_PyTorch

Dependencies

  • Python 3.8.5
  • PyTorch 1.7.1
  • numpy
  • Pillow
  • torchvision
  • scipy

Expriments

Reults of the SIDD validation dataset

To train and evaluate the model directly please visit SIDD website and download the original Noisy sRGB data and Ground-truth sRGB data from SIDD Validation Data and Ground Truth and place them in data/SIDD_Small_sRGB_Only folder.

Pretrained model

Download config.json and model_best.pth from this link and save them in models/CVF_SID/SIDD_Val/ folder.

NOTE: The pretrained model is updated at March. 9th 2022.

You can now go to src folder and test our CVF-SID by:

python test.py --device 0 --config ../models/CVF_SID/SIDD_Val/config.json --resume ../models/CVF_SID/SIDD_Val/model_best.pth

or you can train it by yourself as follows:

python train.py --device 0 --config config_SIDD_Val.json --tag SIDD_Val

Citation

If you find our code or paper useful, please consider citing:

@inproceedings{Neshatavar2022CVFSIDCM,
  title={CVF-SID: Cyclic multi-Variate Function for Self-Supervised Image Denoising by Disentangling Noise from Image},
  author={Reyhaneh Neshatavar and Mohsen Yavartanoo and Sanghyun Son and Kyoung Mu Lee},
  booktitle={IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year={2022}
}