Dataset: HRSTEM Images of Defective and Non-Defective Quasi-Periodic Materials

This is the image dataset and model used to produce the results reported in the following publication: <br> Dennler, N., Foncubierta-Rodriguez, A., Neupert, T., Sousa, M. (2021). <em>Learning-based defect recognition for quasi-periodic HRSTEM images</em>. Micron, 146(July 2020), 103069. https://doi.org/10.1016/j.micron.2021.103069<br> <br> For questions, please correspond with N. Dennler (n.dennler2<strong> </strong>at<strong> </strong>herts.ac.uk) or with M. Sousa (sou at zurich.ibm.com). <strong>hrstem_defects_dataset.zip:</strong> These are the images and labels used to develop and test the algorithm proposed in the above-mentioned publication. They correspond to high resolution scanning transmission electron microscopy images obtained for various III-V films, namely InP, GaAs, InGaAs and InAlGaAs using a JEOL ARM200F microscope. The raw images have been converted in .tif format with the GMS 3 program from Digital Micrograph. The labels have been created by a microscopy expert. Black: main crystal symmetry (non-defective). Gray: secondary crystal symmetry (symmetry defect). White: blurred (amorphous region or beam defect)<br> <br> <strong>vgg16.zip: </strong>The trained neural network model as well as a detailed description of the training/testing dataset that was used to achieve the results reported in the above-mentioned publication.