PromptSTEM: Attentional Deep Learning Accelerates Quantification of Heterogeneous Catalysts from Electron Microscopy

This codebase provides a generalizable method for automated image analysis of supported nanocalysts in transmission electron microscopy, including single-atom catalysts, sub-nano clusters, and nanoparticles.

Installation

Prerequisites

• PyTorch
• OpenCV

Quick command

pip install -r requirements.txt

Model Checkpoint

• vit_b: ViT-B SAM model.

Usage

• 1. Train and predict segmentation models:

python train.py

Data

The datasets used in this study are all publicly available:

• HAADF-STEM of PtSn@Al2O3 BF-TEM of Au@ZSM5: EMcopilot.
• HAADF-STEM of Pt@NC: AtomDetection_ACSTEM.
• BF-TEM of Pd@C : nNPipe.

Citation

If you find our code or data useful in your research, please cite our paper:

@misc{yuan2025FASTCat,
      title={Deep Learning Enabled Single-Shot STEM Imaging for Ultra-Fast Identification of Supported Catalysts}, 
      author={Wenhao Yuan and Fengqi You},
      year={2025},
}