SA-FARI: Open Video Dataset
ApproachSolutionsImpactMediaConsultingAboutContactDonate
ApproachSolutionsImpactMediaEngageAboutContact
##### CONSERVATION X LABS & META PRESENT:
SA-FARI: Open Video Dataset
10,000+ meticulously annotated videos. 99 species.
One open resource to transform AI for conservation.From Raw Footage to Research-Ready Data, Built Using Meta Segment Anything Model 3
What's Inside:
11,000+
high-quality camera trap videos
99
species across diverse ecoystems
100%
annotated, with segmentation masks and tracklets
| | \# Videos | Duration (min) | \# Species | \# Masklets | \# Annotations(boxes & masks) | \# Video–species pairs(incl. negatives) | \# Sampling Locations(independent sites) | | ----- | --------- | -------------- | ---------- | ----------- | ----------------------------- | --------------------------------------- | ---------------------------------------- | | Train | 10,776 | 2,545 | 91 | 15,141 | 880,361 | 31,282 | 650 | | Test | 833 | 202 | 83 | 1,083 | 62,341 | 2,322 | 91 | | Total | 11,609 | 2,747 | 99 | 16,224 | 942,702 | 33,604 | 741 |
Above: The two panels show the number of videos per species category, broken down by data split. The distribution follows a long-tailed pattern typical of real-world wildlife datasets, with a few dominant species and many rarely observed ones. Notably, several species, such as the Saki monkey, appear only in the test set, reflecting the natural open-world setting of camera trap deployments.
A paper on SA-FARI is available on arXiv. The paper details the technical foundations of the SA-FARI dataset—how it was built, how each video was annotated, and what makes it the largest and most diverse multi-animal tracking dataset ever assembled. It explains the data collection and annotation pipeline, provides full dataset statistics, and presents benchmark results showing how SA-FARI improves the performance of state-of-the-art wildlife detection, segmentation, and tracking models.
Built by Conservation X Labs, Meta, and a global coalition of research and conservation organizations.
This collaboration combines Conservation X Labs’ on-the-ground innovation in wildlife monitoring with Meta's expertise in large-scale computer vision datasets. It also utilizes footage generously contributed by five additional partners:
- Osa Conservation, which works to build biodiversity resilience throughout southwestern Costa Rica.
- Los Amigos Biological Station in Peru, dedicated to advancing research and education related to biodiversity and climate adaptation in the Andes-Amazon region.
- Pan African Programme, which studies chimpanzee populations understand ecological drivers of behavioral diversity and evaluate human cultural evolution.
- The Institute for Game and Wildlife Research, born with the mission of generating and disseminating scientific knowledge that allows maintaining a balance between hunting and conservation.
- Instituto Mixto de Investigación en Biodiversidad, an ecological research group that aims to assess emerging threats to wildlife and ecosystems through collaborative research.
- Ecologists & conservationists seeking scalable monitoring methods
- Policymakers & NGOs accelerating biodiversity protection through better data
- AI researchers developing cutting-edge detection, segmentation, and tracking algorithms
By making this resource freely available, we aim to accelerate breakthroughs in both AI research and real-world conservation.
The SA-FARI dataset and its accompanying publication were made possible only through the dedication of countless researchers, engineers, and conservationists, all of whom we recognize and thank in these additional acknowledgments.
A New Frontier for Wildlife AI
Large-scale ecological monitoring is entering the video era. Camera traps capture complex interactions, subtle behaviors, and environmental dynamics impossible to see in still images. But until now, AI progress has been slowed by a critical gap: the lack of robust, open, annotated video datasets. This dataset changes that — unlocking faster, deeper, and more accurate analysis of wildlife populations, behavior, and biodiversity health. The dataset is ideal for:
Open. Free. Global.
We believe the future of biodiversity monitoring depends on shared tools and open science. This dataset is available to researchers, conservationists, and innovators worldwide—no paywalls, no restrictions.