Auto-Labeling Example Service

August 31, 2026 ยท View on GitHub

This package is a template. Copy it when adding a new annotation service and replace the placeholder implementation with product logic.

A service composes one or more task packages into a runnable annotation pipeline. Services depend on core and their task packages, and register an entrypoint so make run can discover them.

Run

make run SCRIPT=example-service:main
make run SCRIPT=example-service:main \
  ARGS='--input-file payloads/simple.jsonl'
make build IMAGE=example-service:build

Each input is a standard DataEntry containing media_path and data_path. The template creates the scene skeleton and writes:

sidecars/example_task.json
sidecars/example_task_manifest.json
sidecars/pipeline_state.json

The template container includes the checksum-pinned, restricted input-only FFmpeg build for core media telemetry. The example task does not decode or encode media, so the image does not install OpenCV or PyAV.