This python package implements a surrogate model to approximately replicate the Monte-Carlo simulations performed to simulate scanning electron microscopy imaging. Our model accepts three-dimensional microstructure representations of porous materials in the form of lists of primitives. It converts them to a specific data representation suitable for a neural network. A convolutional architecture generates two-dimensional electron microscopy images in a single forward pass. The model performs well on arbitrary microstructures like systems of cubes, even though it was trained on structures consisting of spheres and cylinders only.
The method is described in detail in this publication:
[https://openreview.net/attachment?id=SwO84a6yA5&name=pdf]
If you use our package, please cite this paper:
@inproceedings{pub15015,
author = { Dahmen, Tim and Rottmayer, Niklas and Kronenberger, Markus and Schladitz, Katja and Redenbach, Claudia },
title = {A Neural Model for High-Performance Scanning Electron Microscopy Image Simulation of Porous Materials},
booktitle = {CVPR Workshop on Synthetic Data for Computer Vision (SynData4CV-2024)},
address = {Seattle, OR, United States},
year = {2024},
month = {6},
publisher = {CFV}
}