FathomNet Plankton Data in FiftyOne: Embeddings and CLIP

FathomNet is an open-source, expert-curated ocean imagery database created by MBARI and collaborators to support machine learning for marine science, containing tens of thousands of images, over 175,000 labeled localizations, and more than 2,200 concepts as of its 2022 launch. A self-contained Jupyter notebook demonstrates how to load 200 plankton image crops from two FathomNet collections into the FiftyOne dataset-exploration platform using the FathomNet Python client, with all processing able to run on a laptop. The demonstration uses CLIP embeddings and 2D visualization to reveal visually similar taxonomic clusters while showing that broad labels such as detritus and generic diatoms overlap more widely with named species. It also applies similarity search, zero-shot CLIP classification, interactive confusion-matrix evaluation, and uniqueness scoring to inspect classification errors, identify near-duplicate images, assess label quality, and find data suitable for pruning. The authors suggest extending the approach with fine-tuned models, larger detection datasets, geospatial analysis, and community feedback on potential mislabeled examples.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.