Underwater Computer Vision for Ecologists: A Framework for Curating Image Training Datasets for Object Detection
Talen Rimmer, Colin Bates, Declan McIntosh, Tom Zhang, Alexandra Branzan Albu, Francis Juanes · 2026
Read Underwater Computer Vision for Ecologists: A Framework for Curating Image Training Datasets for Object Detection by Talen Rimmer, Colin Bates, Declan McIntosh, Tom Zhang, Alexandra Branzan Albu, Francis Juanes (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13611200.txt.
Description
Rimmer and five co-authors laid out a ten-step framework in twenty twenty-six for using computer vision to watch life under the sea. Marine ecosystems are changing fast, and hand-checking video does not scale, yet ecologists often lack the labeled images that detection models need.
As a test case, the team gathered over nine thousand hours of video from kelp farms and reference sites off British Columbia. Their steps mix unsupervised and supervised tools to build a labeled training set with less hand work. The effort produced over two hundred forty-one thousand labels across fifty-four animal groupings. A standard detection model then scored well overall, best on clear-cut species and worst on blurry broad groups. More labels meant better scores in a retraining test. The authors advise aiming labels at fine groupings where the payoff is largest.
Catalog record
| CREATOR | Talen Rimmer, Colin Bates, Declan McIntosh, Tom Zhang, Alexandra Branzan Albu, Francis Juanes |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 73 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |