Reproduce “PARIS: Part-level Reconstruction and Motion Analysis for Articulated Objects” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jiayi Liu, Ali Mahdavi-Amiri, Manolis Savva
PARIS, developed by Simon Fraser University, introduces a self-supervised framework for part-level reconstruction and motion analysis of articulated objects, inferring their geometry, appearance, and articulation parameters solely from multi-view RGB images captured in two static states.
Key insights
- Object geometry and motion are tightly intertwined; leveraging this interdependence enables joint learning of both aspects without explicit 3D or semantic supervision.
- Representing a movable part in a 'virtual' canonical intermediate state simplifies learning symmetrical transformations and improves the robustness of part decoupling from sparse observations.
- An entropy-based regularization term applied to the opacity contribution of the mobile field can effectively enforce clean semantic-level part segmentation without explicit part labels.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.