Computer vision is increasingly used to process large volumes of animal video from camera traps, drones, and direct observation. Most existing tools focus on detecting individual actions, while interactions—key to understanding social behavior—remain harder to annotate because available software typically supports either behavioral labeling without localization or localization without interaction capture.
SiLVi is an open-source labeling tool that integrates both functions. Researchers can annotate behaviors and interactions directly in video, generating structured outputs suitable for training and validating computer vision models. The authors position it as a bridge between behavioral ecology and computer vision, enabling more fine-grained automated analysis.
Although developed primarily for animal behavior, the paper notes SiLVi could also be useful for annotating human interactions in videos that require dynamic scene graphs. The software, along with documentation and download instructions, is available online.