EMcopilot: Your label-free copilot for automated electron microscopy image analysis

Training and Code Details

For detailed training procedures and source code, please refer to our GitHub repository:
EMcopilot GitHub Repository

Usage

1. Train and predict segmentation models:
   • 00_01_sam_binary_masking.py - Generates coarse masks using the SAM model.
2. Generate and analyze synthetic masks:
   • 02_01_sam_mask_analysis.py - Analyzes SAM mask properties and extracts morphology prior.
   • 02_02_random_mask_generate.py - Generates synthetic masks by augmenting existing masks.
3. Generate and evaluate images:
   • 03_01_p2p_train.py - Trains a Pix2Pix model for mask-to-EMimage translation.
   • 04_01_pix2pix_predict.py - Runs inference using the trained Pix2Pix model.
4. Domain Adaptation:
   • 05_01_domain_adaptation.py - Applies domain adaptation, including noise and contrast augmentation.
5. UNet++ Training and Inference:
   • 06_01_unet++_train.py - Trains a CBAM-enhanced UNet++ model for segmentation.
   • 07_01_unet++_predict.py - Performs inference using the trained UNet++ model.
6. Analyze DM4 microscopy data:
   • 08_01_in_situ_analysis.py - Analyze HAADF-STEM images of supported nanoparticles in real time.

Installation

Install required packages using:

pip install -r requirements.txt

Citation

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

@misc{yuan2024deepgenerativemodelsassistedautomated,
      title={Deep Generative Models-Assisted Automated Labeling for Electron Microscopy Images Segmentation}, 
      author={Wenhao Yuan and Bingqing Yao and Shengdong Tan and Fengqi You and Qian He},
      year={2024},
      eprint={2407.19544},
      archivePrefix={arXiv},
      primaryClass={cond-mat.mtrl-sci},
      url={https://arxiv.org/abs/2407.19544}, 
}

Copyright | 2025 Qian's Lab@NUS