AtomVision: A machine vision library for atomistic images

TL;DR

AtomVision is a comprehensive machine vision library designed for atomistic images, enabling dataset generation, classification, super-resolution, and integration with microscopy tools for materials research.

Contribution

This work introduces AtomVision, a versatile library that combines multiple machine learning techniques for analyzing atomistic microscopy images, which was not previously available as an integrated tool.

Findings

Generated a dataset of 10,000 materials images.

Compared CNN and GNN models for lattice classification.

Developed a U-Net for atom-background segmentation.

Abstract

Computer vision techniques have immense potential for materials design applications. In this work, we introduce an integrated and general-purpose AtomVision library that can be used to generate, curate scanning tunneling microscopy (STM) and scanning transmission electron microscopy (STEM) datasets and apply machine learning techniques. To demonstrate the applicability of this library, we 1) generate and curate an atomistic image dataset of about 10000 materials, 2) develop and compare convolutional and graph neural network models to classify the Bravais lattices, 3) develop fully convolutional neural network using U-Net architecture to pixelwise classify atom vs background, 4) use generative adversarial network for super-resolution, 5) curate a natural language processing based image dataset using open-access arXiv dataset, and 6) integrate the computational framework with experimental microscopy tools. AtomVision library is available at https://github.com/usnistgov/atomvision.