Generalization to Complex Clothing and Canonical-Space Design

Extend AvatarDynamizer’s generalization to more complex clothing and identify a suitable canonical space, potentially based on a unified two-dimensional representation, to improve robustness to garments with extreme non-rigid deformation.

Background

AvatarDynamizer currently relies on motion tracking and geometry registration, which the paper notes can be less robust for garments undergoing extreme non-rigid deformation. The method also represents humans using a template-based pipeline, motivating investigation of a more suitable canonical space for handling complex clothing and improving generalization across garment types.

The authors explicitly leave both extending generalization to more complex clothing and identifying a suitable canonical space—possibly a unified 2D representation—as future work.