Reproduce “PRISM: Probabilistic Representation for Integrated Shape Modeling and Generation” 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
Lei Cheng, Mahdi Saleh, Qing Cheng, Lu Sang, Hongli Xu, Daniel Cremers, Federico Tombari
PRISM introduces a framework for compositional 3D shape generation and manipulation by integrating part-level Statistical Shape Models (SSMs) and Gaussian Mixture Models (GMMs) with categorical diffusion. This method generates diverse, high-quality 3D shapes with varying part configurations, achieving improved coverage and lower matching distances compared to prior art in unconditional generation, and superior performance in single-view and text-conditioned reconstruction tasks.
Key insights
- Integrating part-level statistical shape models with categorical diffusion effectively bridges the gap between high-fidelity full-shape generation and fine-grained part-level control, while accommodating varying topologies and part counts.
- The compositional representation and statistically grounded framework enable enhanced controllability and interpretability in 3D shape generation, allowing for operations like part completion, mixing, and editing.
- The framework facilitates unsupervised learning of part correspondences, reducing the need for extensive manual annotations and improving scalability for complex 3D object datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.