m3-download

A Python-based toolkit for downloading and manipulating object detection annotations from MBARI's M3 (MBARI Media Management) system.

Overview

m3-download provides utilities to:

• Download images and extract object localizations from the M3 database in Pascal VOC format
• Convert between annotation formats (YOLO, Pascal VOC, JSON)
• Process and manipulate annotation files (filtering, remapping concepts, adding taxonomy)
• Perform quality control and analysis on annotation data

These tools are primarily designed for preparing training data for machine learning object detection models.

Key Features

• Data Acquisition: Query and download images and annotations from M3
• Format Conversion: Convert between Pascal VOC and YOLO annotation formats
• Data Processing:
  • Filter annotations by concept
  • Remap concept labels
  • Add taxonomic information
  • Correct image dimensions
  • Find overlapping localizations
• Dataset Analysis: Count localizations by concept

Getting Started

• See Installation for setup instructions
• Check Usage for an overview of available commands
• Explore the documentation for each subcommand to learn specific functionality

Contact

Author: Kevin Barnard, kbarnard@mbari.org

---

© 2025 Monterey Bay Aquarium Research Institute (MBARI)