Nonblind Image Deconvolution via Leveraging Model Uncertainty in An Untrained Deep Neural Network

In this repository we provide the official implementation of "Nonblind Image Deconvolution via Leveraging Model Uncertainty".

General Information

• Codename: Model Uncertainty NID (IJCV 2022)
• Writers: Mingqin Chen ([email protected]); Yuhui Quan ([email protected]); Tongyao Pang ([email protected]); Hui Ji ([email protected])
• Institute: School of Computer Science and Engineering, South China University of Technology; Department of Mathematics, National University of Singapore

For more information please see:

• [paper]
• [supmat]
• [website]

Requirements

Here is the list of libraries you need to install to execute the code:

• Python 3.6
• PyTorch 1.6.0
• scikit-image
• scipy
• cv2 (opencv for python)

All of them can be installed via conda (anaconda), e.g.

conda install scikit-image

How to Execute Demo

1. Run the demo code in demo_gaussian_noise.py for NID in the presence of AWGN.
2. Run the demo code in demo_poisson_noise.py for NID in the presence of Poisson noise.
3. Run the demo code in demo_real_world.py for NID in Real-World Cases.
4. Run the demo code in demo_kohler.py for NID with inaccurate kernels.
5. Run get_psnr_ssim_list.m in Matlab for quantitative results.

Citation

@Article{chen2022nonblind,
  author    = {Chen, Mingqin and Quan, Yuhui and Pang, Tongyao and Ji, Hui},
  title     = {Nonblind Image Deconvolution via Leveraging Model Uncertainty in An Untrained Deep Neural Network},
  journal   = {International Journal of Computer Vision},
  year      = {2020},
  publisher = {Springer}
}

Contacts

For questions, please send an email to [email protected]