Reproduce “Superquadrics Revisited: Learning 3D Shape Parsing beyond Cuboids” 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
Despoina Paschalidou, Ali Osman Ulusoy, Andreas Geiger
This paper presents a learning-based method for 3D shape parsing that utilizes superquadric primitives, demonstrating enhanced expressiveness compared to cuboids. It introduces an analytical solution for the Chamfer distance, enabling more efficient and stable training for unsupervised part decomposition.
Key insights
- Superquadrics offer continuous parameterization and a unified framework to represent diverse geometric forms, overcoming the expressiveness limitations of simpler primitives like cuboids.
- An analytical formulation for Chamfer distance can significantly reduce gradient variance and computational cost in primitive fitting, making deep learning models for structured 3D representations practical.
- Unsupervised learning with reconstruction and parsimony losses can enable a neural network to decompose complex 3D shapes into semantically consistent superquadric parts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.