Reproduce “UCSG-Net -- Unsupervised Discovering of Constructive Solid Geometry Tree” 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
Kacper Kania, Maciej Zięba, Tomasz Kajdanowicz
Researchers at Wrocław University of Science and Technology introduce UCSG-NET, the first method that automatically discovers Constructive Solid Geometry (CSG) parse trees from 3D shapes without explicit supervision. The model generates interpretable, parametric representations, achieving competitive reconstruction accuracy and offering direct applicability in CAD software.
Key insights
- Unsupervised discovery of a shape's underlying CSG tree structure, including primitives and Boolean operations, is achievable via a differentiable network.
- Approximating discrete operations like occupancy conversion and operand selection with differentiable functions, guided by learnable temperature parameters, enables end-to-end training.
- A two-stage training strategy effectively balances reconstruction accuracy with the generation of unambiguous and interpretable CSG parse trees.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.