Reproduce “Mesh BDF: Barycentric Dominance Field for 3D Native 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

Gaochao Song, Haohan Weng, Luo Zhang, Zibo Zhao, Shenghua Gao

This work introduces the Barycentric Dominance Field (BDF), a continuous surface representation for 3D mesh topology, enabling the use of diffusion models for native mesh generation. The BDF method generates high-quality, production-ready meshes with precise topology, outperforming autoregressive models in scalability and robustness, and facilitates unified generation of both geometry and PBR textures.

Key insights

  • The BDF effectively bridges the gap between discrete mesh topology and continuous generative models, allowing the application of powerful diffusion models to native mesh generation.
  • Defining BDF as the maximum of barycentric coordinates ensures C0 and Lipschitz continuity, providing a stable and learnable signal for neural networks.
  • A decoupled VAE architecture for BDF, separate from shape and material VAEs, is crucial for preventing latent space entanglement and maintaining precise vertex accuracy.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.