Q-MAGIC: Quad-only Mesh Autoregressive Generation via Singularity-Aware Circular Relative Offset Tokenization

Key points

• Autoregressive models have proven effective for triangular mesh generation by transforming it into predicting the next face, converting the non-differentiable connectivity prediction into a sequence of token predictions.1
• We first introduce a tailored serialization scheme for quad meshes, which requires at least two new vertices for the next face and is less flexible than triangles due to strict wiring constraints.1
• To meet these requirements and reduce serialization length, we propose a singularity-aware serialization that prioritizes wiring from singular points, leading to more coherent traversals.1
• Additionally, we propose a circular relative offset tokenization that encodes geometry mainly via local relative offsets.1

Sources (1)

• [1]Q-MAGIC: Quad-only Mesh Autoregressive Generation via Singularity-Aware Circular Relative Offset Tokenization
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 08:28 AM

  To address this, we propose Q-MAGIC, a novel autoregressive model for quad mesh generation.

  Autoregressive models have proven effective for triangular mesh generation by transforming it into predicting the next face, converting the non-differentiable connectivity prediction into a sequence of token predictions.

Extractive summary: sentences quoted from the sources.