.png){kind=icon}
.png){kind=icon}
Return
Self-Supervised Hyperbolic Spectro-Temporal Graph Convolution Network for Early 3-D Behavior Prediction
.png){kind=icon}
OA
AI
DOI：10.1109/TCDS.2025.3561422.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
3-D human behavior is a highly nonlinear spatiotemporal interaction process. Therefore, early behavior prediction is a challenging task, especially prediction with low observation rates in unsupervised mode. To this end, we propose a novel self-supervised early 3-D behavior prediction framework that learns graph structures on hyperbolic manifold. First, we employ the sequence construction of multidynamic key information to enlarge the key details of spatiotemporal behavior sequences, addressing the high redundancy between frames of spatiotemporal interaction. Second, for capturing dependencies among long-distance joints, we explore a unique graph Laplacian on hyperbolic manifold to perceive the subtle local difference within frames. Finally, we leverage the learned spatiotemporal features under different observation rates for progressive contrast, forming self-supervised signals. This facilitates the extraction of more discriminative global and local spatiotemporal information from early behavior sequences in unsupervised mode. Extensive experiments on three behavior datasets have demonstrated the superiority of our approach at low to medium observation rates.
Keywords:
Early 3-D behavior prediction
hyperbolic manifold
self-supervised learning
spatiotemporal interaction
Journal
IF:
4.9
Papers:
1.0K
Citations:
3.5K