Four open-source AI models deployed to automate wildlife species classification from camera-trap images across Africa, North America, Australia, and South America

Hacker News · May 9, 2026

AI Summary

•DeepForestVision, developed by Muséum National d'Histoire Naturelle with 6 partner organizations, classifies species in images and videos from camera traps in African tropical forests. It can be run through AddaxAI without programming knowledge or downloaded from GitHub.
•The Ohio State University–Columbus Zoo model achieves 83.6% overall accuracy on a held-out test set of 118,554 images from 1,169 camera trap locations in Ohio, USA, across 46 species classes. It supports taxonomic fallback, allowing predictions to be assigned to higher taxonomic levels when confidence is below a user-defined threshold.
•Australian Wildlife Conservancy's Model AWC135 classifies 135 labels of Australian species from 33 source locations, with training data from ~1.1 million images. It achieves a macro average F1 score of 95% on the top 108 species and 91% across all 135 species on the held-out test set.
•Microsoft AI For Good Lab's model, trained on 41,904 images across 36 labeled genera from the Magdalena Medio region in Colombia, achieves 92% average classification accuracy within a high-confidence subset, with only 10% of detected animal objects requiring human validation after filtering.

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.