WildlifeBench’s Camera-Trap Model Comparison

Field notes

The benchmark separates exact-species accuracy from macro F1 and an adjusted taxonomic score, so readers can distinguish exact matches from answers given at a broader rank. Filters cover animal group, day or infrared recording, curated difficulty, and source continent; the page cautions that subgroup comparisons are descriptive rather than causal. It also notes that local models BioCLIP and SpeciesNet have no API charge, while their hardware and energy costs were not measured.