PtyINR: Ptychographic Implicit Neural Representation

PtyINR is a deep learning framework for joint reconstruction of complex-valued objects and illumination probes in X-ray ptychography. It leverages implicit neural representations and physics-informed optimization to enable high-quality reconstructions under challenging conditions such as limited scan overlap and short exposure time. This repository provides the main scripts, utilities, and interactive notebooks to run PtyINR on both simulated and real experimental datasets.

---

Installation

1. Clone the repository:

   git clone https://github.com/TISGroup/PtyINR.git
   cd PtyINR

2. Install the dependencies:

   pip install -r requirements.txt

3. Install the tiny-cuda-nn:

   cd PtyINR/tiny-cuda-nn/bindings/torch
   pip install .
   

Usage

1. Input Data Preparation

Your input should be in an .h5 file format containing the following datasets:

Header	Description
diffamp	3D array of diffraction amplitudes, without FFT shift, with shape: (N, H, W) — N: number of diffraction patterns — H, W: height and width of each pattern
points	2D array of scan positions with shape: (2, N) — First row: x-coordinates — Second row: y-coordinates — Units: micrometers (µm)
lambda_nm	Wavelength of the incoming X-ray (in nanometers)
ccd_pixel_um	Pixel size of the detector (in micrometers)
z_m	Distance from the sample to the detector (in meters)

Ensure that the data is collected (or simulated) in transmission mode of X-ray.
For simulated data generation, refer to: PtyINR/data_simulation_and_evaluation.py

---

2. Running Ptychographic Reconstruction

PtyINR supports two workflows:

a. Jupyter Notebooks (Recommended for Exploration & Visualization)

• Configure parameters in parameters.py
• interactive_rec_simulated_data.ipynb — For simulated experiments
• interactive_rec_real_data.ipynb — For real experimental data

b. Python Script (Recommended for Batch or Headless Runs)

• Configure parameters in parameters.py
• Run the script from the terminal:

python main.py

For guidance on how to tune PtyINR's hyperparameters (e.g., learning rates, loss weights), refer to: notes_for_parameters.md

---

Repository Structure

PtyINR/  
├── data/                          # Input datasets
├── demo/
│   ├── interactive_rec_real_data.ipynb    # Notebook for reconstructing real data(user data)
│   ├── interactive_rec_simulated_data.ipynb  # Notebook for reconstructing simulated data
├── doc/                           # Documentations on PtyINR  
│   ├── notes_for_parameters.md    # Guidance on how to tune the hyperparameters in PtyINR  
├── PtyINR/                         # Utility functions  
│   ├── siren.py             # Defines MLP-based architectures for object neural representations  
│   ├── forward.py                 # Implements the forward ptychographic propagation model  
│   ├── data_simulation_and_evaluation.py   # Tools for simulating ptychographic measurements and result evaluation
│   ├── train.py        # Training loop and optimization utilities  
│   ├── tiny-cuda-nn/             # Our probe neural network backbone, modified to use float precision
├── main.py                        # Main script for training  
├── parameters.py                  # Parameter configurations  
├── requirements.txt               # Python dependencies  
└── README.md                      

Citation

If you find our work useful in your research, please cite:

@article{li2025ptyinr,
author = {Li, Tingyou and Xu, Zixin and Gao, Zirui and Yan, Hanfei and Huang, Xiaojing and Li, Jizhou},
title = {Learning neural representations for X-ray ptychography reconstruction with unknown probes},
booktitle = {arXiv},
year={2025}
}

Acknowledgements

We gratefully acknowledge the contributions of the following open-source projects, which have significantly inspired and supported the development of this work:

• Instant Neural Graphics Primitives with a Multiresolution Hash Encoding

• Implicit Neural Representations with Periodic Activation Functions (SIREN)

• PtychoNN: Deep learning of ptychographic imaging

• PtyPy - Ptychography Reconstruction for Python

Their contributions to the deep learning and computational imaging communities have been invaluable in the development of PtyINR.