Generative structure zone diagrams (gSZD)

This repository contains the code for the publication "Mastering processing-microstructure complexity through the prediction of thin film structure zone diagrams by generative machine learning models" arxiv 1910.09468. It contains a Jupyter Notebook for a conditional generative adversarial network (cGAN) that is trained on scanning electron microscopy (SEM) surface microstructure images. The folder Image descriptor files contains process parameters and chemical composition of each Cr-Al-O-N SEM surface image. Process parameters and chemical composition are the conditional parameters for the cGAN. Addtional models are a CNN classifier for microstructure classification and a variational autoencoder (VAE). Model weights for the cGAN generator and the microstructure classifier CNN are provided. The image data set is available through Harvard Dataverse.