MSPABDN

This repo contains the KERAS implementation of "Blind Gaussian Deep Denoiser Network using Multi-Scale Pixel Attention"

Run Experiments

To test for blind Gray denoising using MSPABDN write:

python Test_gray.py

The resultant images will be stored in 'Test_Results/Gray/'

To test for blind Color denoising using MSPABDN write:

python Test_color.py

The resultant images will be stored in 'Test_Results/Color/'

Image wise PSNR & SSIM as well as Average PSNR & Average SSIM for the whole image database is also displayed in the console as output.

Train MSPABDN gray denoising network

To train the MSPABDN gray denoising network, first download the BSD400 dataset and save this dataset inside the main folder of this project. Then generate the training data using:

python Generate_Patches_Gray.py

This will save the training patch 'img_clean_pats.npy' in the folder 'trainingPatch/'

Then run the MSPABDN model file using:

python MSPABDN_Gray.py

This will save the 'MSPABDN_Gray.h5' file in the folder 'Pretrained_models/'.

Train MSPABDN color denoising network

To train the MSPABDN color denoising network, first download the CBSD432 dataset and save this dataset inside the main folder of this project. Then generate the training data using:

python Generate_Patches_Color.py

This will save the training patch 'img_clean_pats.npy' in the folder 'trainingPatch/'

Then run the MSPABDN model file using:

python MSPABDN_Color.py

This will save the 'MSPABDN_Color.h5' file in the folder 'Pretrained_models/'.

Citation

@inproceedings{thakur2022blind, title={Blind Gaussian Deep Denoiser Network using Multi-Scale Pixel Attention}, author={Thakur, Ramesh Kumar and Maji, Suman Kumar}, booktitle={2022 IEEE International Conference on Visual Communications and Image Processing (VCIP)}, pages={1--5}, year={2022}, organization={IEEE} }