The research behind SA-FARI, the world's largest open dataset of annotated camera trap wildlife videos, has been selected as an Award Candidate at the Conference on Computer Vision and Pattern Recognition (CVPR), widely regarded as the leading venue for visual AI. The paper (available here) will be presented on Saturday, June 6 in Denver by Dídac Surís Coll-Vinent, a research scientist at Meta.
Approximately 0.06% of total submissions to CVPR are nominated for an award. This recognition is a notable milestone for a dataset built first and foremost for conservation. CVPR's Award Candidate designation singles out only a handful of the accepted papers from among thousands of submissions (16,092 in 2026). Seeing a wildlife-monitoring benchmark earn that distinction is a strong signal that the problems facing conservationists are on the radar of the broader computer vision research community.
SA-FARI was released in November 2025 through a collaboration between Conservation X Labs (CXL) and Meta, and is freely available at conservationxlabs.com/sa-fari. The dataset gathers more than 10,000 expertly annotated clips spanning nearly 100 species across multiple continents. What sets it apart is its detail: every frame carries bounding boxes and segmentation masks, making SA-FARI the first resource of its scale to bring that level of annotation to such a diverse body of camera trap footage.
That combination of breadth and precision addresses a long-standing bottleneck. AI for conservation has leaned heavily on still images, which capture a moment but miss the movement — the behaviors, interactions, and subtle indicators of animal health that only video reveals. Without robust, annotated video benchmarks to train and test against, progress has been slower than the urgency of the work demands. SA-FARI changes that.
The dataset is powered by Meta's Segment Anything Model 3 (SAM 3), a computer vision model capable of detecting, outlining, and tracking objects across every frame of a video. Footage was contributed by partners working at the front lines of conservation, including Osa Conservation in Costa Rica, the Los Amigos Biological Station in Peru, and collaborators across Central Africa.
SA-FARI is available to the global community with no restrictions, reflecting a shared commitment to open science and to accelerating the pace of conservation innovation. You can read paper on SA-FARI here, which will be the focus of Saturday's presentation in Denver.
