Reasoning for Implicit Hand-Object Contact Estimation from Monocular RGB
This project explores a new approach to hand-object interaction by predicting dense physical contact directly on the surfaces of the hand and object. Given only a monocular RGB image and a canonical 3D object template, we will develop a model for interaction-aware reasoning that infers object-space contact without relying on explicit poses. The project draws on current directions in canonical-space prediction, physically grounded scene understanding, and 3D correspondence learning to reason about visual evidence of contact regions on the object surface.
