Unlike single-person lifters that process one skeleton at a time, our baseline encodes the entire "group." We argue that the pose of Person A provides strong priors for Person B if they are interacting.
Precise GPS updates that show family members on a private map. family threesixtyp
Family360 addresses these gaps. We utilize 360-degree panoramic cameras to capture rich household dynamics, allowing for the reconstruction of ground-truth 3D poses via multi-view geometry. This paper details the data collection pipeline, annotation methodology, and a thorough evaluation of contemporary models. Unlike single-person lifters that process one skeleton at
Understanding human behavior requires accurate 3D pose estimation. While datasets like Human3.6M and MuPoTS have driven progress, they suffer from limitations: subjects are often isolated, or interactions are staged. In real-world settings—specifically family environments—humans constantly interact: hugging, carrying children, or sitting closely together. These scenarios introduce: We utilize 360-degree panoramic cameras to capture rich
The plan was to spend 365 days together, documenting every moment of their lives. No exceptions, no alone time, no secrets. They would film, photograph, and journal their experiences, exploring the depths of their relationships and themselves.