Skip to main content

Installation

Recommended runtime prerequisites:

  • Python 3.11
  • Valid model paths, local service endpoints, or API backends
  • Node.js >= 20 if you also plan to build this documentation site

The repository already provides a unified installation script. That should be your default starting point.

cd <PROJECT_ROOT>
./scripts/install.sh annotate

This command tries uv first and falls back to other package managers if needed. For the full environment matrix, see environments/README.md.

Install the CLI wrapper​

If you want to call autopipeline directly instead of typing python -m src.cli.autopipeline every time, run:

cd <PROJECT_ROOT>
./scripts/install_autopipeline.sh
autopipeline --help

Manual installation​

If you want explicit control over the virtual environment, you can install with uv:

cd <PROJECT_ROOT>
uv sync --frozen --extra annotate
uv pip install -r environments/requirements/non_pypi/annotate.txt

If you prefer pip or conda, the same script supports both:

./scripts/install.sh annotate --manager pip
./scripts/install.sh annotate --manager conda --env-name annotate

User configuration​

AutoPipeline depends heavily on configs/pipelines/user_config.yaml. At minimum, verify these two sections:

  • client_config Model service endpoints, API keys, ports, timeouts, and retry settings.
  • model_paths Local model paths for CLIP, DINO, SAM, ArcFace, Depth Anything, and related assets.

Recommended practice:

  1. Do not commit real API keys to a public repository.
  2. Maintain a private local override or a machine-specific copy of the config.
  3. Make sure every model_paths entry points to a real location on the current machine.

Additional model preparation​

It is a good idea to download the InsightFace assets before your first multi-process run, so workers do not race on first-time initialization:

wget https://github.com/deepinsight/insightface/releases/download/v0.7/buffalo_l.zip

Verify the installation​

At minimum, validate these three entry points:

autopipeline --help
python -m src.cli.autopipeline --help
python -c "import src.autopipeline"

Documentation site note​

If you are maintaining the documentation site itself rather than the runtime, Docusaurus requires Node.js >= 20. The host environment in this workspace still defaults to Node 16, so local builds should use a Node 20 runner.

If you want the shortest path to a successful runtime command, continue with Quickstart.