/prDeep

This package contains the code to run prDeep; a noise robust phase retrieval algorithm based on deep neural networks.

Primary LanguageMatlabOtherNOASSERTION

Original release date : 2/28/18

Reference #1 : "prDeep: Robust Phase Retrieval with Flexible Deep Neural Networks"

Authors : Christopher A. Metzler, Philip Schniter, Ashok Veeraraghavan, Richard G. Baraniuk

Questions/suggestions/comments: chris.metzler@rice.edu

Primary Contents

Scripts:

  • PR_demo.m: Demonstrates phase retrieval from noisy coded diffraction and fourier measurements using prDeep.

Functions:

  • prDeep.m: Implementation of prDeep
  • HIO.m: Basic implementation of HIO algorithm

Auxiliary functions:

  • disambig2Drfft.m: Phaseless 2D fourier transforms have phase and translation ambiguities. This function accounts for them.

Packages

This download includes trained DnCNN denoiser weights for various noise levels.

Dependencies

D-AMP Toolbox (https://github.com/ricedsp/D-AMP_Toolbox) must be on your path. MatConvNet (http://www.vlfeat.org/matconvnet/) must be compiled and on your path. FASTA (https://github.com/tomgoldstein/fasta-matlab/) must be on your path.

Installation

MatConvNet must be compiled before it can be used.