Skip to content

Code for our IEEE TCSVT Paper: Lightweight Modules for Efficient Deep Learning based Image Restoration

Notifications You must be signed in to change notification settings

avisekiit/TCSVT-LightWeight-CNNs

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 

Repository files navigation

TCSVT-LightWeight-CNNs

Code for our IEEE TCSVT Paper: Lightweight Modules for Efficient Deep Learning based Image Restoration

Authors: Avisek Lahiri*, Sourav Bairagya*, Sutanu Bera, Siddhant Haldar, Prabir Kumar Biswas
(* equal contribution)

  1. Paper Link: https://arxiv.org/abs/2007.05835
  2. IEEE Early Access Link: https://ieeexplore.ieee.org/document/9134805

Key Points from Paper

  • Paper provides re-usable modules to be plugged and played to compress a given CNN
  • Select any favourite full-scale baseline for low-level vision applications
  • Replace 3X3 conv by LIST layer
  • Replace dilated conv layer GSAT layer
  • Achieve efficient up/down-sample with Bilinear SubSampling followed by LIST layer

TensorflowExamples

This contains the basic proposed modules in Tensorflow
TensorflowExamples/basicModules.py contains the proposed LIST, GSAT modules
It also contains the framework for LIST based up/down-sampling in a CNN

PytorchExamples

It contains the basic proposed modules in Pytorch

  • block.py has the implementatin of LIST module based DnCNN denoising framework
  • Denoising_demo.ipynb is a notebook to reflect our training/inference setup for DnCNN experiments

Cover Picture

About

Code for our IEEE TCSVT Paper: Lightweight Modules for Efficient Deep Learning based Image Restoration

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published