Reproduce “Rethinking 3D Shape Generation: Diffusion over Superquadrics” 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
Zhiyang Liu, Wanze Li, Yuwei Wu, Chengran Yuan, Jiawei Sun, Rui Zheng, Marcelo H Ang Jr
Researchers from the National University of Singapore and Singapore University of Technology and Design introduced Diffusion over Superquadrics (DoSs), a method that generates 3D shapes by diffusing a compact set of superquadric primitives. This approach achieves competitive 3D shape generation performance while drastically reducing computational costs and enabling direct structural editing and geometric-aware design.
Key insights
- Diffusing over a compact, explicit set of geometric primitives (superquadrics) can drastically reduce the state space for 3D generation, improving efficiency while maintaining competitive quality.
- Explicitly modeling the existence of primitives and using a stable 6D rotation representation are crucial for robustly learning the distribution of superquadric parameters.
- A deterministic canonical ordering of primitives before diffusion helps the model learn consistent structural relationships by mitigating permutation ambiguity inherent in unordered sets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.