Skip to content

arsenal9971/DeNSE

master
Switch branches/tags

Name already in use

A tag already exists with the provided branch name. Many Git commands accept both tag and branch names, so creating this branch may cause unexpected behavior. Are you sure you want to create this branch?
Code

Latest commit

 

Git stats

Files

Permalink
Failed to load latest commit information.
Type
Name
Latest commit message
Commit time
 
 
 
 
 
 
 
 
 
 

DeNSE: Deep Network Shearlet Edge Extractor

By H. Andrade-Loarca, G. Kutyniok, O. Öktem, P. Petersen

Architecture

License

DeNSE is released under the MIT License (refer to the LICENSE file for details).

Contents

  1. Introduction
  2. Citation
  3. Installation
  4. Usage
  5. Contact

Introduction

This repository contains the entire pipline (including data preprocessing, training, testing, evaluation and visualization) for DeNSE.

DeNSE is a recently proposed learning framework towards wavefront set detection in 2D-arrays, in other words, DeNSE computes the edges of an image and its orientations for further applications. This method uses the optimal edge representation in images provided by Shearlets and the highly specilized and accurate classification capabilities of deep convolutional neural networks. For more details, please refere to the arXiv technical report.

Using a simple 4-layered CNN DeNSE achieves state-of-the-art edge orientation extraction performance on the Semantic Boundaries Dataset and the Berkeley Segmentation Dataset, as well as other toy dataset with ellipses and parallelograms. It presents also capabilities to detect singularities for different degrees of regularity.

This method can be used for different applications in image processing and computer vision (e.g. edge/corner detection and tracking) as well as inverse problems regularization (e.g. Wavefront set reconstrucion in Computed Tomography).

Citation

If you find DeNSE useful in your research, please consider to cite the following papers:

@inproceedings{andrade2019dense, 
  title={Extraction of digital wavefront sets using applied harmonic analysis and deep neural networks}, 
  author={Andrade-Loarca, Hector, Kutyiniok, Gitta, Öktem, Ozan, Petersen, Philipp},
  booktitle={arXiv preprint: arXiv:1901.01388}, 
  year={2019}
}

Installation

You can install all the dependencies by using the conda local enviroment file.

conda env create -f environment.yaml
conda activate dense

Usage

The training and evaluation of the model for each of the used dataset is presented in the results folder.

Contact

Hector Andrade-Loarca

Questions can also be left as issues in the repository. We will be happy to answer them.

About

Code for "Extraction of digital wavefront sets using applied harmonic analysis and deep neural networks"

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published