DeepForestVisionV2: Ecology-Driven Taxonomy Expansion for Camera-Trap Monitoring in African Tropical Forests

Researchers have released DeepForestVisionV2, an expanded artificial-intelligence model that nearly doubles the number of species it can identify in camera-trap images from African tropical forests, according to a preprint posted on arXiv [1]. The new model builds on the original DeepForestVision tool, which was designed for closed-canopy, ground-level forest interiors and used a 35-class prediction space [1]. That taxonomy proved too coarse when camera deployments encountered arboreal primates, birds, semi-aquatic taxa, or human-associated confounders such as livestock [1]. DeepForestVisionV2 expands the prediction space to 64 classes — 61 animal classes plus human, vehicle, and blank — to address three recurrent deployment gradients: vertical stratification, scene openness, and anthropogenic interfaces [1]. The model was trained on 1,535,010 photographs and 243,354 videos drawn from multi-country African tropical-forest projects [1]. It retains the same offline workflow as its predecessor, which the authors note is the only open tool providing a matched offline workflow for both photographs and videos in this domain [1]. On a cross-country cropped-photo validation set, DeepForestVisionV2 reached 0.86 accuracy, 0.82 macro-F1, and 0.81 balanced accuracy [1]. The researchers also tested the model on three held-out video benchmarks from Uganda that span the targeted gradients [1]. In forest-interior videos, the number of identified taxa rose from 22 to 29, while at riverbanks the count increased from 4 to 9 [1]. At park edges — a setting that combines human activity with wildlife — accuracy climbed from 0.62 to 0.86 and false alarms dropped from 11 to zero [1]. The work arrives as camera-trap monitoring in African tropical forests increasingly moves beyond closed-canopy interiors to riverbanks, clearings, and park edges [1]. The preprint was posted on arXiv, an open-access repository of electronic preprints that, as of November 2024, receives about 24,000 submissions per month and has surpassed two million articles [7]. The paper appears under the Computer Vision and Pattern Recognition category and is accompanied by arXivLabs integrations, a framework that allows community collaborators to develop and share experimental tools directly on the arXiv website [5][6].