Deblurring with neural networks

This project is made for our signal class at Université Gustave Eiffel.

Requirements for the project

• Each project requires reading and understanding.

• Each project requires some coding (not only using some existing toolbox).

• Each project requires a testing phase of the methods on your own data.

Presentation

The duration of the presentation is 15/20 minutes.

~10-15 slides with the following wireframe :

• Title / name
• Presentation of the problem
• Proposed methods
• Theoretical analysis
• Numerical findings
• Critics
• Conclusion / perspective

Report

• You must bring a printed version with your to give me before the presentation. This is a strict deadline.
• Not less than 4 pages and up to 12 pages (A4, 11pt font, single column), including roughly one page of bibliography.
• Should be similar to a (short) scientific article.
• When presenting the numerics, give all parameters, so that the results are reproducible.
• It should not be a rewriting of the original article. You should only report about what you have done and only explain the theory that is relevant to explain the numerics you have done.
• Suggestion of structure for the report:
  • Abstract:
    • What problem(s) is studied ?
    • Why is it relevant ?
    • What solution(s) is proposed ?
    • Which contributions (theory, numerics, etc) ?
  • Introduction :
    • Presentation of the problem(s).
    • Previous works (at least a few citations). If relevant, include things that you have seen during the lectures.
    • Contributions. Why is the studied method different/better/worse/etc. than existing previous works.
  • Main body (~10 pages) :
    • Presentation of the method(s).
    • Theoretical guarantees.
    • Numerics.
  • Conclusion and perspective (~1 page)
    • Summary of the result obtained: pros and cons (limitation, problems, error in the articles, etc)
    • Possible improvement/extension
  • Bibliography (~1 page)

Get dataset

wget https://drive.google.com/uc?export=download&id=1P0mqAANuR-iGzfeYtKYczTH_Nd8-JoCW

Research papers

Name of the article	Authors	Link
Scale-recurrent Network for Deep Image Deblurring	Xin Tao, Hongyun Gao, Yi Wang, Xiaoyong Shen, Jue Wang and Jiaya Jia	https://arxiv.org/pdf/1802.01770.pdf
A Review of Convolutional Neural Networks for Inverse Problems in Imaging	Michael T. McCann, Kyong Hwan Jin and Michael Unser	https://arxiv.org/pdf/1710.04011.pdf
A New Deep Learning Method for Image Deblurring in Optical Microscopic Systems	Huangxuan Zhao, Ziwen Ke, Ningbo Chen, Ke Li, Lidai Wang, Xiaojing Gong, Wei Zheng, Liang Song, Zhicheng Liu, Dong Liang and Chengbo Liu	https://arxiv.org/ftp/arxiv/papers/1910/1910.03928.pdf