Reproduce “TopGen: Learning Structural Layouts and Cross-Fields for Quadrilateral Mesh 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
Yuguang Chen, Xinhai Liu, Xiangyu Zhu, Yiling Zhu, Zhuo Chen, Dongyu Zhang, Chunchao Guo
TopGen converts dense triangle meshes into quadrilateral meshes by predicting structural lines and a cross-field together. Structural lines mark important surface paths, while the cross-field guides quad-edge directions; a QuadWild-inspired backend then performs remeshing. On the AI-generated Robot, TopGen reports a Chamfer Distance of 3.300 × 10⁻⁵ versus 4.044 × 10⁻⁵ for QuadWild, but at different output resolutions.
Key insights
- Jointly predicting both explicit structural lines and implicit cross-fields is crucial for simultaneously achieving high geometric fidelity at boundaries and maintaining topological regularity within the mesh.
- A geometry-aware encoder with dual cross-attention, processing point clouds sampled adaptively around sharp features, significantly enhances robustness to varied and non-manifold input geometries.
- Mimicking professional modeling workflows (layout-first, generation-second) through learned explicit structural guidance can yield automatically generated quad meshes that meet industry-level quality and editability standards.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.