Reproduce “SQuadGen: Generating Simple Quad Layouts via Chart Distance Fields” 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
Youkang Kong, Yang Liu, Yue Dong, Xin Tong, Heung-Yeung Shum
SQuadGen generates simple quad layouts by learning continuous chart distance fields conditioned on input geometry rather than predicting mesh connectivity directly. Its pipeline extracts layouts from the fields, while test-time regularization improves loop simplicity. On ABC1k, it scores 0.78 versus 0.69 for QuadRemesher, with fewer charts, but has higher Hausdorff distance and does not guarantee all-quad outputs.
Key insights
- Transforming discrete mesh connectivity into continuous Chart Distance Fields enables deep learning models to effectively learn and generate complex quad layout topologies.
- A data-driven, generative approach using latent diffusion models can synthesize artist-friendly quad layouts that are robust, coherent, and suitable for loop-based editing operations.
- Quantitative metrics for "loop simplicity" are essential for objectively evaluating the editability of quad mesh layouts and for curating high-quality datasets for training generative models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.