DAMA: Disentangled Body-Anchored Gaussians for Controllable Multi-Layered Avatars

Researchers introduced DAMA, a Gaussian-based 3D avatar method for reconstructing clothed humans from multi-view images.

DAMA anchors Gaussians to body geometry, producing cleaner garment boundaries and physically plausible clothing layers.

The method lets users control garment stacking and reorder layers for more editable avatar reconstruction.

It reports state-of-the-art results on the 4D-DRESS dataset, with improved realism and lower penetration in clothed human models.