From Extrinsic to Intrinsic: Geodesic-Guided Representation Learning for 3D Geometric Data

Researchers introduced PRISM, a geodesic-guided pretraining framework for 3D representation learning.

PRISM learns intrinsic surface geometry by recovering geodesic distances, rather than relying only on extrinsic shape cues or semantics.

The method adds topology constraints and a two-stage training strategy to reduce distance imbalance during learning.

It performs strongly on geodesic prediction, shape recognition, surface parameterization, and non-rigid correspondence.