This repository provides an automated pipeline for analyzing 4D-STEM datasets using YOLOv8n.
The workflow enables end-to-end processing of large-scale 4D-STEM datasets
for phase identification, orientation mapping (coming soon!), and strain analysis.
Phase mapping of complex phase-transformed Ti-50Nb alloy using object detection-based pattern recognition.
Strain mapping of Si/SiGe multilayers demonstrating coherent lattice mismatch analysis.
Supported file formats:
- Thermo Fisher Scientific:
.emi,.xml(EMPAD) - GATAN:
.dm3,.dm4 - Dectris:
.h5 - NanoMegas:
.blo - Direct Electron:
.de5 - Standard:
.h5,.hdf5
Python ≥ 3.9 is required.
We recommend creating a new virtual environment:
conda create -n tempo4d python=3.9
conda activate tempo4d⚡ Install PyTorch (Recommended First)
If you have a CUDA-capable GPU, install a CUDA-compatible version of PyTorch before installing tempo4d.
👉 Install PyTorch
📦 Install tempo4d
pip install tempo4d
This will install all required dependencies, including:
- PyQt5
- pyqtgraph
- OpenCV
- matplotlib
- Ultralytics (for YOLOv8)
- rosettasciio[all] (for TEM file support)
Please also see the tempo4d_demo.ipynb notebook in the demo folder.
Download example data from GATAN
@InProceedings{Genc_2025_ICCV,
author = {Genc, Arda and Silverstein, Ravit},
title = {Neural Object Detection for 4D-STEM: High-Throughput Sub-Pixel Electron Diffraction Pattern Recognition},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops},
month = {October},
year = {2025},
pages = {3565-3575}
}