/powcodgen

GPU-accelerated generation of diffraction patterns from the PowCod database

Primary LanguagePythonMIT LicenseMIT

python build and test

powcodgen

The code in this repo aims to generate semi-realistic diffraction patterns from the PowCod database for machine learning applications.

Once you have obtained a copy of the database, you can filter it as desired, using the notebook in this repo.

Several method of data augmentation have been implemented:

  • variable unit cells to simulate temperature changes (whilst maintaining crystal-system)
  • variable peak intensities to simulate preferred orientation (March-Dollase)
  • variable Gaussian, Lorentzian and FCJ axial divergence contributions to full Voigt peaks
  • variable background profile (Chebyshev polynomials)
  • variable background noise
  • zero-point errors

See this post for more information