coolsyn2000 / NeuralNetwork-PhaseRetrieval

Data-driven and untrained phase retrieval network for MNIST datasets

PhaseRetrieval-MNIST

[Pytorch] Phase retrieval network based on Pix2Pix and Untrained Neural Network

Architecture

• Pix2Pix

  

• Untrained Neural Network

  

Result

• run python run_predict.py

  

• run python ./untrained/phase_retrieval.py

  

Train

• Test Environment: Python 3.9, Pytorch 2.1, CUDA 11.8

• Test Hardware: CPU i7-11800H, GPU Nvidia RTX 3060 Laptop, RAM DDR4 16G

• For training with MNIST, just run python run_train.py in terminal. Results (generated_images, state_dict() , loss_curve) while training will be saved in ./result/*

• For running with untrained method, run python ./untrained/phase_retrieval.py.

Data

MNIST and its autocorrelation generated by fn_generate_dataset.py

• MNIST_autocorr_test.pt:

  shape:(1000, 1, 128, 128) type: float

• MNIST_autocorr_train.pt:

  shape:(9000, 1, 128, 128) type: float