Machine Learning for Next-Generation Transmission Electron Microscopy (TEM)

TEM-ParticlesDB is an open repository of Transmission Electron Microscopy (TEM) nanoparticle images and associated segmentation data that have been automatically segmented using machine learning to identify particle size, shape and morphology across a range of substrates. Each entry includes the original experimental TEM images, the synthetic training data used to develop the segmentation models, and the resulting segmented particle outlines. The synthetic data is generated using the TEMPOS (Transmission Electron Microscopy Particle Outline Segmentation) software and used to train segmentation models via transfer learning, before being applied to the experimental images.

TEMPOS is software used to segment TEM nanoparticle images by generating synthetic training data and training segmentation models that underpin the datasets in TEM ParticlesDB. It provides a graphical interface for creating synthetic TEM images, applying contrast transfer functions, training segmentation models, and processing experimental images to produce particle outlines and measurements. For users who want to reproduce or extend the workflow, the repository also includes Jupyter notebooks for segmenting TEM videos, with or without particle tracking, and scripts for running the full method outside the GUI. These materials allow researchers to explore how the segmentation models were trained, adapt the workflow to their own data, or generate additional synthetic datasets for further model development.

This guidance resource provides full documentation, supported by walkthrough videos, on how to use TEMPOS, the graphical interface for generating synthetic TEM data, applying CTF, training Detectron2 segmentation models, processing experimental images, and exporting data locally with Datasette. It includes an overview of the software, instructions for setting up the required environment, step by step guidance for running the tool, installation notes for Windows users, and options for more advanced execution.