Maintained by UCI DeepEM Lab


Journal Reference: TEMImageNet training library and AtomSegNet deep-learning models for high-precision atom segmentation, localization, denoising, and deblurring of atomic resolution images, by Ruoqian Lin, Rui Zhang, Chunyang Wang, Xiao-Qing Yang & Huolin L. Xin, Scientific Reports, 11, 5386 (2021) https://doi.org/10.1038/s41598-021-84499-w

The TEM ImageNet Project was seeded by the Battery500 Consortium in conjunction with Dr. Xiao-Qing Yang. It is now maintained by Prof. Huolin Xin's DeepEM Lab at UC Irvine under the program of Office of Basic Energy Sciences of the U.S. Department of Energy, under award no. DE-SC0021204 (program manager Dr. Jane Zhu).

We have ported the AtomSegNet models to Matlab. We will deploy these models a universal App that can be installed on any OS without you needing to worry about installing Python and dependencies.

TEM ImageNet v1.3 is now released to the public and the datasets are available to download from Github.

Our AtomSegNet tool is used by the broader community. Check out the work that has utilized our tool.

Data is available for searching and browsing on TEMImageNet.org

The deep-learning models and the AtomSegNet app are available at https://github.com/xinhuolin/AtomSegNet

Synthetic datasets

Realistic synthetic ADF-STEM images that can fool experienced electron microscopists.

Ground Truth Labels

Include ground-truth labels for intensity-preserving super-resolution processing, atomic-column Gaussian mapping, noise reduction, denoising+background removal, atomic-column segmentation.

Also include position labels for training region proposal networks.


The atomic coordinates and the imaging condition of each synthetic image is provided for reproducibility and other types of learning tasks.