sleap-docker-container

Docker image for running sleap-nn training and tracking workflows.

This container is designed primarily for running inside Pennsieve processors, where the platform provides INPUT_DIR and OUTPUT_DIR. It is also set up so the same behavior can be tested locally with docker run.

The entrypoint currently supports two modes:

• train
• track

Run all commands below from inside sleap-docker-container/.

Build

docker build -t sleap-uv .

Directory Layout

The local examples below assume this layout:

data/
  input/
    sleap_config.yaml
    train.pkg.slp
    val.pkg.slp
  models/
    my_first_model/
  output/

Train

This command runs:

sleap-nn train --config /app/input/sleap_config.yaml

Use:

docker run --rm -it \
  -e RUN_MODE=train \
  -e INPUT_DIR=/app/input \
  -e OUTPUT_DIR=/app/output \
  -e CONFIG=sleap_config.yaml \
  -v "$(pwd)/data/input":/app/input \
  -v "$(pwd)/data/":/app/output \
  sleap-uv

Track

This command runs:

sleap-nn track \
  --data_path /app/input/val.pkg.slp \
  --model_paths /app/models/my_first_model/ \
  -o /app/output/val.predictions.slp

Use:

docker run --rm -it \
  -e RUN_MODE=track \
  -e INPUT_DIR=/app/input \
  -e OUTPUT_DIR=/app/output \
  -e MODEL_INPUT_DIR=/app/models \
  -e DATA_PATH=val.pkg.slp \
  -e MODEL_PATHS=my_first_model/ \
  -e O=val.predictions.slp \
  -v "$(pwd)/data/input":/app/input \
  -v "$(pwd)/data/models":/app/models \
  -v "$(pwd)/data/output":/app/output \
  sleap-uv

Environment Variables

• RUN_MODE: train or track
• INPUT_DIR: base directory for config and input data
• OUTPUT_DIR: base directory for outputs and -o
• CONFIG: config file path relative to INPUT_DIR for train
• DATA_PATH: input file path relative to INPUT_DIR for track
• MODEL_PATHS: model directory path relative to MODEL_INPUT_DIR for track
• MODEL_INPUT_DIR: optional model base directory for track; defaults to INPUT_DIR
• O: output file path relative to OUTPUT_DIR for track

Pennsieve Note

Pennsieve provides INPUT_DIR and OUTPUT_DIR. For Pennsieve, the preferred setup is to make all required tracking inputs available under INPUT_DIR. MODEL_INPUT_DIR is mainly useful for local Docker runs when models live outside the main input directory.