Code from my Bachelor Thesis. It is focused on creating CNNs and evaluating predictions of microstructure properties for different parameters.
Due to the variety of libraries and the high computational demand, the notebooks were written for Google Colab. At each point in the notebooks suitable adaptations for running notebooks locally are marked. Using the same file structure of the repository in drive can be useful in reducing set up effort.
The notebooks were used to do a variety of things. randomMatrixGen can be used to generate more data. It is based on code by Tim Duka.
The helper_functions folder contains several libraries and folders with classes and functions used for the thesis. The models folder contains templates for models that were successfully used in this thesis.
In the repository for each image size a data set with contrast ratio 100 (or 0.01) and its binary and two-point variants is provided. On request I can provide the remaining contrast ratio data sets and other variants.
If something does not work, if it is too messy, or if you want more data contact me or leave something in the issues section :)