Developer Guidelines for MegaDetector Pipeline

Current Configuration

  • Model: The MegaDetector model currently used is md_v5b.0.0.pt, stored in ml_public_models.
  • Scripts: The main processing script is cameratraps_pipeline.py, executed in multiple phases.
  • Environment: The pipeline uses the megadetector-env (version 4) with required dependencies.

Project-Specific Configuration

Each organization or project has a unique JSON configuration file that customizes the pipeline. When creating or updating a pipeline for a new project, ensure:

  • A new JSON config file is created.
  • Project-specific parameters (e.g., datastore paths, compute resources) are correctly referenced.

How to Modify the Pipeline

Changing the Model

  1. Upload the new model to the ml_public_models datastore.
  2. Update the model path in the pipeline parameters (e.g., megadetector_naturalstate_rbp_pipeline_parameters.json).

Updating the Script

  1. Modify the cameratraps_pipeline.py script.
  2. Ensure the script is correctly referenced in the pipeline configuration.

Changing Compute Resources

  • If you need to modify the compute clusters (GPU or CPU):
    1. Go to the pipeline settings in AML Studio.
    2. Update the compute parameter in the relevant pipeline steps.

Testing Changes

  • Always test pipeline modifications in a development environment before deploying to production.