🐳 Docker Support

Aegear provides a ready-to-use Docker image to simplify deployment and training in cloud environments or on systems without a configured Python environment.

---

📦 Public Docker Image

The official Aegear Docker image is hosted on Docker Hub:

ljubobratovicrelja/aegear:latest

Pull the image with:

docker pull ljubobratovicrelja/aegear:latest

---

🚀 Running the Image

Launch a container with GPU support (if available):

docker run --gpus all -it ljubobratovicrelja/aegear:latest

To mount a local directory for accessing datasets or saving models:

docker run --gpus all -it -v /path/to/data:/workspace/data ljubobratovicrelja/aegear:latest

This will mount /path/to/data on your host to /workspace/data inside the container.

---

🛠️ Training with Docker

The recommended workflow for training models is to use the CLI-based training script via the container entrypoint. The image includes all dependencies and uses run_training.sh to configure and launch training.

Example:

docker run --gpus all -e MODEL_TYPE=efficient_unet -e DATA_MANIFEST=/workspace/data/manifest.json -e MODEL_DIR=/workspace/models/unet -e CHECKPOINT_DIR=/workspace/models/unet/checkpoints ljubobratovicrelja/aegear:latest

Set additional environment variables to customize training (see docker/README.md for details).

---

⚠️ Notebook-based Training (Deprecated)

Notebook-based training (training_unet.ipynb, training_siamese.ipynb) is still available for development, but is deprecated and will be removed in future releases. Please use the CLI training workflow for new projects.