A Stereo Camera Dataset of Annotated Budgerigar Flight Trajectories for Studying Avoidance Behaviors in Multi-agent Scenarios

The Bird Stereo 3D Dataset is a structured, multi-modal dataset designed to support advanced research in multi-object tracking, stereo vision-based reconstruction, and behavioral analysis of flying birds under both isolated and interactive scenarios. This dataset integrates synchronized stereo image annotations, structured metadata, and reconstructed three-dimensional trajectories, offering a comprehensive benchmark for both biological and computational studies. The dataset has been meticulously curated to represent a variety of flight configurations, ranging from single-bird motion to complex multi-agent collision avoidance scenarios involving up to six birds simultaneously. Each component of the dataset—from raw annotations to processed outputs—has been organized in a hierarchical and reproducible format to ensure accessibility, consistency, and scalability.