MegaDetector Pipeline on Azure Machine Learning (AML)

Project Overview

The MegaDetector Pipeline processes camera trap images to detect animals, humans, and vehicles using a pre-trained MegaDetector model. The pipeline is fully implemented on Azure Machine Learning (AML) and uses existing compute clusters, datastores, and environments within the same Azure subscription.

Current Pipeline Details

• Pipeline Name: MegaDetector-NaturalState-RBP-Pipeline
• Datastores:
  • landing_kutuma_hashed: For raw input image data.
  • ml_public_models: For storing the MegaDetector model.
  • bronze_camera_trap: For intermediate processed results.
  • bronze_megadetector: For final detection results.

Key Features

• Object Detection: Detects animals, humans, and vehicles in camera trap images.
• Phased Execution: Pipeline executes in multiple phases to handle large image datasets.
• Batch Processing: Efficient image processing using GPU/CPU clusters in Azure.
• Results Storage: Outputs are stored in Azure Blob Storage for easy access.

Documentation

For detailed instructions, see the documentation files in the docs/ folder:

• Pipeline Architecture
• Usage Instructions
• Developer Guidelines
• Troubleshooting