Structured Random Model for Fast and Robust Phase Retrieval

This repository contains the source code for the paper submission "Structured Random Model for Fast and Robust Phase Retrieval" to ICASSP2025.

Overview

Our work presents a novel model for phase retrieval, offering both speed and robustness. The implementation is based on the open-source computational imaging library deepinv.

Repository Structure

• experimental/paper/scripts/: Contains scripts for data generation
• experimental/paper/notebooks/: Includes Jupyter notebooks for figure generation

Getting Started

1. Clone the repository:

   git clone https://github.com/zhiyhucode/structured-random-phase-retrieval.git
   cd structured-random-phase-retrieval

2. Install the required dependencies:

   1. Install uv;
   2. Run uv sync under the root directory of the project;
   3. Optionally, run uv pip install --no-build-isolation fast-hadamard-transform, if you have a GPU;
   4. Activate the environment using source .venv/bin/activate;

3. Navigate to the scripts directory to generate data, e.g.:

   cd experimental/paper/scripts
   python pseudorandom_spectral.py

4. Use the notebooks in experimental/paper/notebooks/ to reproduce the figures from the paper.

Contributing

We welcome contributions to improve the code or extend the research. Please submit a pull request or open an issue for any bugs or feature requests.

License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.

Citation

If you are interested in citing the work, please cite our paper:

@inproceedings{hu2025structured,
  title={Structured Random Model for Fast and Robust Phase Retrieval},
  author={Hu, Zhiyuan and Tachella, Juli{\'a}n and Unser, Michael and Dong, Jonathan},
  booktitle={ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2025},
  organization={IEEE}
}