Solving Phase Retrieval with a Learned Reference (ECCV2020)

This repository is the official implementation of our paper: Solving Phase Retrieval with a Learned Reference

Environment

Experiments conducted under:

• python 3.7
• pytorch 1.31

Datasets: dataset.py

Training

To learn the reference, open Jupyter notebook: Train_abs.ipynb

📋 Some parameters:

• the number of iterations (n_steps)
• the number of training epochs (n_epoch)
• training samples (n_train)
• batchsize (n_batch)
• stepszie (alpha)
• learning rate (lr_u)

Evaluation

To evaluate the learned reference on different datasets, run: Test_mismatch.ipynb

📋 This code includes transformed test images.