Building rearticulable models for 3D articulated objects from multi-view RGB videos

This thesis presents a solution to the novel task of building a 3D rearticulable models from multi-view colored (RGB) videos, which enables to render the articulable object at sampled state, discover the rigid parts, and reconstruct the kinematic structure. The solution is comprised of Gaussian Articulated Implicit Model (GAIM), a representation for dynamic scenes with an articulable object, and a relax-and-project-based optimization framework. Evaluated on the Watch-It-Move (WIM) dataset, the proposed solution achieves on par view synthesis performance as other rendering baselines, and can effectively discover rigid parts and properly reconstruct the kinematic model of the observed articulated object.

Name thesis:degree_name

M.S.

Level thesis:degree_level

Thesis

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2024

Author dc:creator

• Jiang, Wei

Contributors dc:contributor

• Wang, Shenlong

• 3d Articulated Object Understanding
• Gaussian Splatting
• Neural Rendering
• Part Discovery

Statement dc:rights

• Copyright 2024 Wei Jiang

Language dc:language

en, eng

Handle dc:identifier

https://hdl.handle.net/2142/125661

OAI record oai:identifier

https://www.ideals.illinois.edu/oai-pmh?verb=GetRecord&metadataPrefix=etdms&identifier=oai%3Awww.ideals.illinois.edu%3A2142%2F125661