Arti4D: Statistical Analysis and Modeling of the Spatio-temporal Variability in Articulated 4D Shapes

By decoupling bending from stretching, our framework avoids non-plausible artifacts that occur in SRNF-based methods.

A 4D surface is represented as a trajectory in the product space of shape and pose parameters.

Temporal alignment is performed by mapping trajectories to SRVF space and computing optimal time-warping via Dynamic Programming.

Our method accurately recovers temporal alignment even when random temporal diffeomorphisms introduce visible phase shifts.

Our method yields natural deformations along geodesic paths, avoiding unnatural shrinkage artifacts that occur in other methods.

We simultaneously compute the mean motion and optimal time warpings to synchronize motion phases across multiple 4D surfaces.

After temporal co-registration, PCA reveals meaningful principal modes of variation that capture genuine motion variability.

Novel 4D surfaces are generated by sampling along principal modes of variation from the learned statistical model.