A Deep Learning Pipeline for 3D Cell Detection, Tracking, and Lineage Reconstruction in Zebrafish Microscopy Data

• Name: Biohub — Cell Tracking During Development
• Goal: Detect, track, and link cells in 3D microscopy data of developing zebrafish embryos.
• Metrics: Edge Jaccard + Division Jaccard (higher is better).
• Data Format: .zarr 3D volumetric time-series.

git clone https://github.com/your-username/ZebraTrack3D.git
cd ZebraTrack3D
pip install -r requirements.txt

Or with Conda:

conda env create -f environment.yml
conda activate zebratrack3d

Place raw Kaggle data in data/raw/:

kaggle competitions download -c biohub-cell-tracking-during-development
unzip biohub-cell-tracking-during-development.zip -d data/raw/

python scripts/train.py --config configs/params.yaml

python scripts/predict.py --input data/raw/test.zarr --output outputs/predictions/submission.csv

python scripts/evaluate.py --pred outputs/predictions/submission.csv --gt data/processed/ground_truth.csv

python scripts/submit.py --file outputs/predictions/submission.csv --message "3D U-Net + MCF v1"

ZebraTrack3D/
├── configs/          # YAML configuration files (hyperparams, paths)
├── data/             # Raw, processed, and external data
├── notebooks/        # Jupyter notebooks for EDA and prototyping
├── src/              # Core source code (data, models, utils)
│   ├── data/         # Data loading and preprocessing
│   ├── models/       # Detection, tracking, and lineage models
│   └── utils/        # Metrics, visualization, helpers
├── scripts/          # Train, predict, evaluate, submit
├── outputs/          # Saved models, predictions, logs, figures
├── tests/            # Unit and integration tests
└── docs/             # Architecture, setup, and results docs

See docs/architecture.md for a detailed breakdown.

Edit configs/params.yaml to adjust hyperparameters:

model:
  name: unet3d
  in_channels: 1
  out_channels: 2
  features: [32, 64, 128, 256]

training:
  epochs: 100
  batch_size: 2
  lr: 1e-4

1. Fork the repository.
2. Create a feature branch: git checkout -b feature/my-feature.
3. Commit your changes: git commit -m "Add my feature".
4. Push to the branch: git push origin feature/my-feature.
5. Open a Pull Request.

This project is licensed under the MIT License — see LICENSE for details.

• ChanZuckerberg Biohub for the competition data.
• OME-Zarr for the .zarr format.
• PyTorch and the open-source community.