Geometric and Semantic Coupling for Interaction Understanding in 3D Scenes
Segment-Snap ties movable parts, handles, and motion paths together instead of predicting them as separate outputs.
The paper uses learned predictors for broad part surfaces and small handles, then applies geometric priors to constrain motion. Handle locations guide hinge-line selection without training a motion regressor. On Articulate3D validation, that handle guidance raises motion-gated AP from 13.74% to 40.98% under fixed masks and axes. Added handle candidates and part-based class correction lift handle AP to 30.99% with full context. HF Daily Papers' note
The paper uses learned predictors for broad part surfaces and small handles, then applies geometric priors to constrain motion. Handle locations guide hinge-line selection without training a motion regressor. On Articulate3D validation, that handle guidance raises motion-gated AP from 13.74% to 40.98% under fixed masks and axes. Added handle candidates and part-based class correction lift handle AP to 30.99% with full context. HF Daily Papers' note
score 4