Note to Hackathon organizers: The code for our analysis and figures is in the notebooks folder. The atom_gui.py script can be run to test the deep learning pipeline on the included dm3 file.
This project addresses the challenge of defect classification for 2D material STEM datasets. Identifying and classifying defects in scanning transmission electron microscopy (STEM) images is crucial for understanding material properties and improving manufacturing processes. By developing automated machine learning solutions, we can accelerate the analysis of these complex datasets and provide more consistent, reliable defect characterization.
This project focuses on identifying defects in an open-source HAADF-STEM (High-Angle Annular Dark-Field Scanning Transmission Electron Microscopy) dataset from Oak Ridge National Laboratory (ORNL). The dataset contains STEM images of MoWSSe (Molybdenum-Tungsten-Sulfur-Selenium) 2D materials, which may exhibit various structural defects that affect their properties.
A relevant dataset is publicly available from ORNL, containing STEM images of MoWSSe materials:
- Dataset Link: ORNL Dataset
- Content: HAADF-STEM images of MoWSSe 2D materials
- Format: Includes data and accompanying Jupyter notebook
The primary goal of this hackathon project is to develop machine learning models to identify and classify defects reliably in STEM images. This includes:
- Detection of various defect types in the 2D material structure
- Classification of identified defects
- Development of robust, reproducible ML pipelines
- Evaluation of model performance on the dataset
(To be added as the project develops)
# Example dependencies
# - Python 3.x
# - NumPy
# - TensorFlow/PyTorch
# - Jupyter Notebook
# - Other scientific computing libraries(To be added as the project develops)
# Clone the repository
git clone https://github.com/callen350/haadf-stem-hackathon.git
cd haadf-stem-hackathon
# Install dependencies
# pip install -r requirements.txt(To be added as the project develops)
# Example workflow
# 1. Download the dataset
# 2. Run preprocessing scripts
# 3. Train models
# 4. Evaluate results(To be updated as the project develops)
haadf-stem-hackathon/
├── README.md # This file
├── data/ # Dataset and processed data (to be added)
├── notebooks/ # Jupyter notebooks for exploration and analysis (to be added)
├── src/ # Source code for models and utilities (to be added)
└── results/ # Model outputs and evaluation metrics (to be added)
Contributions to improve the defect classification models or extend the project are welcome! Please feel free to submit issues or pull requests.
(To be determined)
- Oak Ridge National Laboratory (ORNL) for providing the open-source HAADF-STEM dataset
- All hackathon participants and contributors
For questions or collaboration opportunities, please open an issue in this repository.