AnyPose: Anytime 3D Human Pose Forecasting via Neural Ordinary Differential Equations
This addresses a crucial need for synchronous real-world human-machine interaction by enabling anytime predictions, representing a novel method for a known bottleneck in the field.
The paper tackles the problem of 3D human pose forecasting at any real-valued time step, which existing methods only handle at preset intervals, by introducing AnyPose, a lightweight continuous-time neural architecture using neural ordinary differential equations; results show it achieves high-performance accuracy and significantly lower computational time on datasets like Human3.6M, AMASS, and 3DPW.
Anytime 3D human pose forecasting is crucial to synchronous real-world human-machine interaction, where the term ``anytime" corresponds to predicting human pose at any real-valued time step. However, to the best of our knowledge, all the existing methods in human pose forecasting perform predictions at preset, discrete time intervals. Therefore, we introduce AnyPose, a lightweight continuous-time neural architecture that models human behavior dynamics with neural ordinary differential equations. We validate our framework on the Human3.6M, AMASS, and 3DPW dataset and conduct a series of comprehensive analyses towards comparison with existing methods and the intersection of human pose and neural ordinary differential equations. Our results demonstrate that AnyPose exhibits high-performance accuracy in predicting future poses and takes significantly lower computational time than traditional methods in solving anytime prediction tasks.