Skip to content
forked from nikopj/DGCN

Deep Graph Convolutional Network for image restoration.

Notifications You must be signed in to change notification settings

RShamshiri/DGCN

 
 

Repository files navigation

DGCN: Deep Graph Convolutional Network for image denoising.

This repository contains a PyTorch implementation of the paper Deep Graph Convoluitonal Image Denoising.

File Descriptions:

  • analyze.py: contains functions for obtaining training curves, test-set performance, etc. given an argument file (i.e. $ python3 analyze.py path/to/args.json)
  • args.json: sample arguments file used to interface with model configurations and checkpoints.
  • data.py: defines datasets and data-loading functions.
  • knn.py: defines functions for peforming K-nearest neighbors with local-masks.
  • net.py: defines GCDN network and submodules.
  • train.py: used like analyze.py, initializes and fits a model to training data given an arguments file.
  • utils.py: defines functions for data pre/post processing, indexing, etc.
  • visual.py: visualization tools written with matplotlib package, such as an interactive receptive field viewer.

Further Improvements:

  • Faster nearest neighbors computations, perhaps with packages such as FAISS.

About

Deep Graph Convolutional Network for image restoration.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • TeX 51.2%
  • Python 46.5%
  • Shell 2.3%