Search papers, labs, and topics across Lattice.
The paper introduces Strips as Tokens (SATO), a novel token ordering strategy for autoregressive transformer-based artist mesh generation that constructs sequences as connected chains of faces encoding UV boundaries. This approach addresses limitations of existing methods by preserving organized edge flow and semantic layout, leading to higher quality meshes. SATO's unified representation allows joint training on triangle and quad mesh data, leveraging triangle data for structural priors and quad data for geometric regularity.
Achieve artist-quality 3D meshes with autoregressive transformers by ditching coordinate-based sorting and patch-based heuristics for a triangle-strip-inspired tokenization that respects UV boundaries.
Recent advancements in autoregressive transformers have demonstrated remarkable potential for generating artist-quality meshes. However, the token ordering strategies employed by existing methods typically fail to meet professional artist standards, where coordinate-based sorting yields inefficiently long sequences, and patch-based heuristics disrupt the continuous edge flow and structural regularity essential for high-quality modeling. To address these limitations, we propose Strips as Tokens (SATO), a novel framework with a token ordering strategy inspired by triangle strips. By constructing the sequence as a connected chain of faces that explicitly encodes UV boundaries, our method naturally preserves the organized edge flow and semantic layout characteristic of artist-created meshes. A key advantage of this formulation is its unified representation, enabling the same token sequence to be decoded into either a triangle or quadrilateral mesh. This flexibility facilitates joint training on both data types: large-scale triangle data provides fundamental structural priors, while high-quality quad data enhances the geometric regularity of the outputs. Extensive experiments demonstrate that SATO consistently outperforms prior methods in terms of geometric quality, structural coherence, and UV segmentation.
CMU ML