Search papers, labs, and topics across Lattice.
This paper introduces the Barycentric Dominance Field (BDF), a novel continuous representation for 3D mesh generation that addresses the limitations of autoregressive (AR) modeling by encoding vertex topological connectivity on triangular mesh surfaces. By transforming discrete mesh topology into a continuous surface signal, BDF enhances the capabilities of diffusion-based generative models, allowing for the generation of higher quality, scalable, and robust native meshes. Experimental results show that BDF outperforms state-of-the-art AR methods in terms of mesh quality and scalability, marking a significant advancement in 3D mesh generation techniques.
BDF enables diffusion models to generate 3D meshes with unprecedented quality and scalability, overcoming the limitations of traditional autoregressive approaches.
Autoregressive (AR) modeling has recently achieved remarkable progress in native 3D mesh generation, largely due to its natural ability to handle variable-length, discrete data structures. However, the inherent constraints of the AR paradigm severely restrict the generated meshes, leading to limited face counts, bounded vertex resolutions, and difficulties in supporting textures. To overcome these bottlenecks, we propose the Barycentric Dominance Field (BDF), a continuous representation defined on triangular mesh surfaces that elegantly encodes vertex topological connectivity. BDF bridges the fundamental gap between discrete mesh topology and continuous diffusion-based generative modeling by transforming connectivity into a continuous surface signal. As an intrinsic mesh property, BDF shares strong similarities with texture maps, enabling its seamless integration into existing 3D diffusion pipelines without requiring architectural modifications. Extensive experiments demonstrate that BDF empowers diffusion models to generate native meshes with significantly higher quality, greater scalability, and stronger robustness compared to state-of-the-art autoregressive methods.
