How Conservation X Labs Is Using Segment Anything Model 3 for Endangered Wildlife Monitoring

Imagine driving down a Florida highway at night and seeing a mountain lion — one of only a few left in the state — dart across the road. Each sighting is a reminder: As Florida’s wild places shrink, so does the chance for these iconic animals to survive. If the mountain lions disappear, the balance of the entire ecosystem breaks down — impacting everything from deer populations to the plants that keep Florida’s forests healthy.

Recognizing the urgent need for proactive and innovative conservation solutions, start-ups like Conservation X Labs (CXL) and government agencies like Florida Fish and Wildlife Conservation Commission (FWC) are using Meta’s Segment Anything Models to bolster on-the-ground conservation expertise.

Our newest model, SAM 3, introduces the ability to detect, segment, and track objects in video using short text phrases. For example, SAM 3 could be prompted with names of animal species such as “green iguana” or “Bezoar goat.” This extends the capabilities of SAM 2, which enables segmenting and tracking objects in videos using clicks and bounding boxes.

“Conservation X Labs tracks hundreds of species around the world to measure their survival and prevent their extinction,” says Dr. Alex Dehgan, co-founder of CXL. “SAM 3 will help us improve that dramatically because, for the first time ever, we will have the ability to automate the description and monitoring of animal behavior.”

Their work illustrates the transformative potential of integrating technological advancements with on-the-ground conservation expertise. Rather than viewing technology and nature as opposing forces, these leaders embrace their synergy, showing how human ingenuity — supported by advancing science — can help build a more resilient planet.

Founded in 2015 by Dr. Alex Dehgan and Dr. Paul Bunje, CXL operates on the premise that open innovation, diverse collaboration, and technology must work in tandem to keep pace with our planet’s rapid biodiversity loss.

CXL facilitates innovation challenges and prizes, leads field programs worldwide, hosts wildlife and data platforms, and develops tools with conservationists to scale their impact. Since its inception, CXL has hosted 20 innovation challenges, provided over $12 million in funding to breakthrough solutions, re-identified more than 299,000 animals, and expanded protected areas by nearly 400,000 acres. Their open innovation programs have resulted in over $577 million in follow-on funding to innovators after working with CXL.

Large-scale ecological monitoring has been digitally transformed by use of the camera trap video, which captures complex and subtle interactions and behaviors that cannot be seen in still images. But until now, there has been a critical shortage of large, publicly accessible, and well-annotated camera trap video datasets.

“The current lack of these resources severely impedes AI advancement in ecological science and conservation,” says Lona Stoll, COO and Innovation Counsel of CXL.

To address this challenge, CXL, Meta, Osa Conservation, Los Amigos Biological Station, Pan African Programme, The Institute for Game and Wildlife Research, and Instituto Mixto de Investigación en Biodiversidad collaborated by combining on-the-ground wildlife monitoring with Meta’s latest Segment Anything Model 3 (SAM 3) to build SA-FARI, an open dataset of research-ready, raw video footage.

The public dataset includes over 10,000 camera trap videos of more than 100 species, meticulously annotated with bounding boxes and segmentation masks for every animal in each frame, preserving identity over time (tracklets). The dataset was made possible by the contribution of video footage from Osa Conservation, which works on the ground, hand-in-hand with local communities, government officials, and global leaders to build biodiversity resilience throughout southern Costa Rica, as well as Los Amigos Biological Station and other conservation organizations.

The video footage is an input to SAM 3. The model can then detect, segment, and track individual animals in camera trap videos, which is essential for non-invasive, large-scale ecological monitoring. These videos provide extensive spatio-temporal data on wildlife populations, distribution, behavior, and disease.

SA-FARI is the largest open source dataset of its kind and an important benchmark for object detection, counting, segmentation, species classification, and tracking in ecology, conservation, and machine learning. It’s also the first public dataset combining high-quality segmentation and tracklet annotations in camera trap videos.

“Having an open dataset allows us to supercharge conservation and have a common basis on which to develop new models for a variety of use cases, from analyzing biodiversity to understanding animal health,” says Dr. Dehgan.

SA-FARI enables the broader AI community to develop innovations for wildlife and biodiversity conservation worldwide. Stoll emphasizes, “This dataset is truly groundbreaking and unique for conservation because of its detail, quality, and scale. Making this freely available globally is a gift.”

FWC began in 1964 as the Wildlife Research Project and evolved into today’s Conservation Commission, a state government agency that manages and protects Florida’s rich natural resources. These include more than 575 wildlife species, over 700 native fish species, 34 million acres of public and private land, and 12,000 square miles of water.

FWC balances serving approximately 20 million residents and millions of visitors who share the land and water with Florida’s wildlife while maintaining biodiversity and ecosystem health for native species’ long-term wellbeing. This balance is crucial, as protecting Florida’s wildlife and wildlands also supports the state’s economy. Wildlife viewing, fishing, and hunting contribute over $15 billion annually and sustain tens of thousands of jobs.

The significance of SA-FARI is very real to FWC, which is partnering with CXL to improve monitoring and protection of the Florida mountain lion, a subspecies of Florida panther and the last surviving Puma subspecies in the eastern United States. Once widespread across the southeastern US, the Florida mountain lion now occupies less than 5% of its historic range, with a single breeding population of about 200 individuals in south Florida.

Recently, FWC observed increased cases of feline leukomyelopathy (FLM), a neurological disease causing limb weakness and paralysis, which can lead to starvation and death. Monitoring this elusive, mostly nocturnal animal is challenging. Traditional methods are labor-intensive and often fail to provide timely, actionable insights, especially critical given the disease’s impact.

To address these challenges, CXL leveraged Meta Segment Anything Model 2 (released in July 2024) to track individual mountain lions in camera trap videos. SAM 2 produces “tracklets” — segmentation masks in video — enabling observations of the same animal over time. This enables behavior detection and real-time identification of diseased animals, making disease management feasible.

The use of SA-FARI and SAM 3, released in November 2025, will build upon the success FWC has seen with SAM 2.

“We will be using SAM 3 to harness the full potential of camera trap videos to identify and track animals and their behaviors in complex environments,” says Dr. Dehgan.

SA-FARI and SAM 3 are an important milestone towards scalable, accurate wildlife monitoring systems that can empower conservationists to make data-driven interventions with speed and precision. Looking ahead, continued advancements in AI and computer vision will further transform ecosystem protection — enabling proactive strategies, broader collaboration, and global conservation impact. These tools pave the way for smarter, more adaptive solutions to safeguarding endangered species and habitats across our planet.