A new AI system called SA-FARI can automatically detect, name and follow individual animals across nearly 100 species in video footage. Built on Meta’s latest vision model, the tool could save wildlife researchers thousands of hours of manual review.
Conservationists and wildlife researchers may soon spend far less time scrubbing through hours of camera trap footage, thanks to a new AI system capable of automatically identifying and tracking individual animals across nearly 100 species — pixel by pixel, frame by frame.
The system, called SA-FARI (Segment Anything in Footage of Animals for Recognition and Identification), was developed by an international consortium led by ConservationX Labs and Meta, with key contributions from a University of Bristol team specializing in Animal Biometrics and AI for Conservation. The paper is being presented June 6 at the Conference on Computer Vision and Pattern Recognition (CVPR) in Denver — widely considered the top venue for visual AI research — where it has been selected as an Award Candidate.
How the Technology Works
SA-FARI is built on Meta’s Segment Anything Model 3 (SAM3), a cutting-edge Vision-Language Model that uses both text and visual prompts to pinpoint, outline and follow objects across images and video. Applied to wildlife footage, the system generates what researchers call “masklets” — precise, frame-by-frame outlines that trace an animal’s exact shape as it moves through a scene. By cleanly separating each animal from its background, the tool creates a foundation for deeper analysis of individual identity and behavior.
To develop and test the system, the project team assembled a dataset of more than 11,000 wildlife videos filmed in natural habitats, all carefully curated and annotated. That dataset is being made freely available to biologists, conservationists and ecologists worldwide, giving researchers anywhere access to a powerful AI toolkit without needing to build one from scratch.
Why It Matters for Conservation
Camera traps generate enormous volumes of footage, but sorting through that content manually is one of the most labor-intensive parts of field research. SA-FARI has the potential to eliminate thousands of hours of that work, freeing scientists to focus on analysis rather than data wrangling.
Otto Brookes, a lecturer in AI and animal biometrics at Bristol, explained that the stakes go beyond convenience.
“The ability to locate animals in space and time is incredibly important for wildlife monitoring – it is a prerequisite for many tasks such as recognising behaviour and distinguishing individuals from one another and ultimately measuring how animals respond to conservation interventions,” Brookes said in a news release.
In other words, tracking animals accurately isn’t just a technical milestone — it’s the foundation for answering some of the most pressing questions in conservation science, from how endangered populations respond to habitat restoration to how individual animals behave under environmental stress.
Tilo Burghardt, a professor of computer vision and animal biometrics at Bristol, framed the project within a broader global mission.
“Global problems require global solutions. Based on the group’s pioneering track record of over 20 years, the University of Bristol is regarded as one of the go-to places for using AI for conservation in the UK and beyond, and is an important part of a growing international community working in this area,” Burghardt said in the news release.
A Collaborative, Interdisciplinary Effort
The scale of SA-FARI reflects just how many disciplines had to come together to make it work. Beyond Bristol and the lead organizations, contributors included teams from the Hasso Plattner Institute, the University of Oviedo, Osa Conservation, the Senckenberg Museum of Natural History, the Max Planck Institute for Evolutionary Anthropology, and Climate Corridors. First author Dante Wasmuht and senior author Didac Suris spearheaded the research effort.
The Bristol team’s recognition at CVPR is also notable for another reason: this marks the second consecutive year the group has earned a prestigious nomination at the conference, signaling the university’s growing international profile in AI-driven conservation research.
Burghardt noted that SA-FARI is designed with extensibility in mind, suggesting future researchers could build on the system to incorporate features such as tracking animal body posture, estimating depth, and generating natural language descriptions of animal behavior.
What This Means for Students and Emerging Researchers
For students studying ecology, wildlife biology, computer science or data science, SA-FARI represents a tangible example of how AI tools are reshaping fieldwork. The freely available dataset and model lower the barrier to entry for student-led research projects, senior theses, or graduate work involving wildlife monitoring. Students interested in conservation technology now have access to a state-of-the-art benchmark and a model trained on real-world conditions — without needing institutional resources to build one.
The paper, “The SA-FARI Dataset: Segment Anything in Footage of Animals for Recognition and Identification,” is available as a preprint.
Source: University of Bristol