Fast-Image-Filters

Fast-Image-Filters copied to clipboard

Fast Image Processing with Fully-Convolutional Networks

PyTorch implementation

Report Bug · Request Feature

Table of Contents

• About the Project
  • Description
• App Demo
• Getting Started
  • Code Setup
  • Data Download and Preparation
  • Training
  • Testing

About The Project

This project is an extension to the previous project here on edge aware filters.

The link to paper is here

Description

This paper shows effectiveness of the following in implementing image processing filters

• Fully Convolutional architecture with dilated filters, that leads to better receptive field for image filters and very smaller models
• Neat trick on Identity weight initialization
• Simple Mean Square Error loss

App Demo

Image Filters

Getting Started

Code Setup

# Clone the repository
$ git clone https://github.com/nrupatunga/Fast-Image-Filters.git

# install all the required repositories
$ cd Fast-Image-Filters
$ pip install -r requirements.txt

# Add current directory to environment
$ cd src
$ source settings.sh

Data Download and Preparation

Since the author has not released the dataset and dataset preparation script, I couldn't share the same here. Please mail [email protected] for the scripts and data. Feel free to message me for any help you need.

Training

$ cd Fast-Image-Filters/src/run/

# Modify the data_dir variable in train.py

# run to train
$ python train.py

Testing

$ cd Fast-Image-Filters/src/run

# Change the model paths in app.py

$ python app.py

Contact

Email: [email protected]