WildTrack | Wildlife detection, segmentation, and tracking from videos

Conservation teams often rely on manual annotation or costly proprietary software to monitor wildlife from camera-trap or drone footage. While tools like MegaDetector and Segment Anything 2 provide powerful components for detection and segmentation, using them together seamlessly requires significant technical setup.

WildTrack was created to bridge this gap — offering a clean, modular, open-source pipeline that conservationists and researchers can use out of the box to generate high-quality masks and track individuals automatically across frames.

WildTrack integrates MegaDetector (for animal detection) and SAM 2 (for segmentation) in an incremental loop. The system runs detection every N frames, adds new animals only when necessary, and propagates existing masks forward using SAM 2’s visual object tracking.

This incremental strategy allows the system to handle long videos efficiently while maintaining temporal consistency and high-quality segmentation masks.

In test runs with camera-trap and safari footage, WildTrack successfully maintained consistent IDs for multiple animals over hundreds of frames, producing research-ready segmentation masks.

By combining two state-of-the-art vision models into a ready-to-use pipeline, WildTrack lowers the technical barrier for wildlife researchers and NGOs who need robust, automated visual tracking. The open-source release allows teams to adapt the system for different species, habitats, or recording setups, enabling reproducible and collaborative conservation research.