MERMAID classifier

PySpacer-based tooling for training and evaluating classifiers for the MERMAID conservation platform.

Code: github.com/data-mermaid/mermaid-classifier
Built: 2024

What it is

Training and evaluation tooling, built on PySpacer, for the classifiers behind MERMAID’s benthic image annotation. This work spanned multiple years in collaboration with UC Davis and MERMAID coral scientists. We worked with CoralNet’s data to create initial classification models based on MERMAID’s classes and priorities.

What I did

• Wrote training and evaluation scripts, with evaluation logged to MLflow.
• Generated self-contained HTML reports from MLflow runs, templated with Jinja2.
• Built a Dockerized CLI for reproducible runs.
• Packaged it for pip install, with optional extras such as PySpacer and JupyterLab, so users install only what they need.
• Exported CoralNet annotations to Parquet for downstream analysis.

For classifier results at scale, see Coral classification at scale.