Photon Limited Non-Blind Deblurring Using Algorithm Unrolling

Pytorch code for Photon Limited Non-Blind Deblurring Using Algorithm Unrolling - published at Transactions on Computational Imaging

Pretrained model here

Instructions

1. Create a local copy of repository using the following commands

   foor@bar:~$ git clone https://github.com/sanghviyashiitb/poisson-deblurring.git
   foor@bar:~$ cd poisson-deblurring
   foor@bar:~/poisson-deblurring$ 
   
   

2. Download the pretrained model into model_zoo from the link here

3. To test the network using synthetic data, run the file

   foo@bar:~/poisson-deblurring$ python3 demo_synthetic.py  
   

   Output:

   

4. Download the zip file containing real dataset into the main directory and unzip using the following command:

   foo@bar:~/poisson-deblurring$ unzip real_data.zip -d data/ 
   

5. To test the network using real data, run the file

   foo@bar:~/poisson-deblurring$ python3 demo_synthetic.py  --idx=11
   

   (Variable idx represents the file index and can be any integer from [0,29] )

   

   Output: PSNR: 29.08, SSIM: 0.696

Training

Before running train.py, add clean images (for example Flickr2K) in the data/training and data/val folders.

Citation

@ARTICLE{9903556,
 author={Sanghvi, Yash and Gnanasambandam, Abhiram and Chan, Stanley H.},
 journal={IEEE Transactions on Computational Imaging}, 
 title={Photon Limited Non-Blind Deblurring Using Algorithm Unrolling}, 
 year={2022},
 volume={8},
 number={},
 pages={851-864},
 doi={10.1109/TCI.2022.3209939}}

Feel free to ask your questions/share your feedback at [email protected]