naveenkumarcm / xrd_denoising Goto Github PK

This repository contains Python source code for training a deep convolutional neural network to denoise experimental low-counting statistics X-ray diffraction data. It provides the neural-network definitions as well as a training pipeline. The expected format of low- and high-count data is TIF. Furthermore, the data is expected to be located in separate folders named "LC" and "HC" for both training and validation data sets. A helper function for converting the provided (training, validation, and test) HDF5 files at Zenodo (https://doi.org/10.5281/zenodo.8237173) to individual TIF files is given in helper_functions.py, named convert_zenodo_hdf5_to_tif().

Required packages (Python 3):

• numpy
• pandas
• pillow
• scipy
• h5py
• tensorflow 2.4.1 (optional but recommended: tensorflow-gpu)

Original work and data:

• J. Oppliger et al., “Weak-signal extraction enabled by deep-neural-network denoising of diffraction data.” arXiv, Sep. 19, 2022. doi: 10.48550/arXiv.2209.09247.
• J. Oppliger et al., “X-ray diffraction dataset for experimental noise filtering.” Zenodo, Jul. 24, 2022. doi: 10.5281/zenodo.8237173.