Precise part segmentation and articulation of 3D models

Develop accurate and robust techniques to precisely segment 3D models into functional parts and correctly articulate each part, moving beyond limited object categories and simple motion types such as prismatic translation, rotation, and revolution.

Background

Part-based representations assign semantic and functional meaning to decomposed components and are widely used for modeling articulated objects. The authors note that current approaches rely on segmented datasets or language-guided reasoning but still struggle with generalizability and functional understanding.

Existing applications are largely confined to specific categories (e.g., furniture) and simple motions, indicating a need for methods that can generalize segmentation and articulation across diverse objects and complex behaviors.