iNaturalist species suggestions from community photos
iNaturalist, a nonprofit where people share photos of plants and animals, trains a computer vision model on photos its community shares, and uses it to suggest what species someone has photographed. Its 2021 model covered over 38,000 kinds of plants, animals and fungi.
What they learned
New training software cut model training from about seven months to two and a half
From public sources: as the photo collection grew from 12 to nearly 21 million images, training slowed. iNaturalist wrote new training software, which it estimated cut training time from about seven months to about two and a half. It also reported that balancing accuracy across regions, and using location information well, were still open problems.
The problem
There are far too few experts to identify every species people photograph, which limits how much biodiversity data the public can produce.
What the AI does
A model looks at a photo and suggests likely species. People in the community then confirm or correct the identification.