FlashMesh: Faster and Better Autoregressive Mesh Synthesis via Structured Speculation

TL;DR

FlashMesh introduces a novel speculative decoding framework for autoregressive 3D mesh generation, significantly speeding up inference while improving fidelity by exploiting structural correlations in mesh data.

Contribution

The paper presents a new predict-correct-verify paradigm and a speculative decoding scheme tailored for autoregressive mesh synthesis, enabling parallel prediction and faster inference.

Findings

Achieves up to 2x speedup over standard models

Improves generation fidelity in mesh synthesis

Effectively leverages structural priors for acceleration

Abstract

Autoregressive models can generate high-quality 3D meshes by sequentially producing vertices and faces, but their token-by-token decoding results in slow inference, limiting practical use in interactive and large-scale applications. We present FlashMesh, a fast and high-fidelity mesh generation framework that rethinks autoregressive decoding through a predict-correct-verify paradigm. The key insight is that mesh tokens exhibit strong structural and geometric correlations that enable confident multi-token speculation. FlashMesh leverages this by introducing a speculative decoding scheme tailored to the commonly used hourglass transformer architecture, enabling parallel prediction across face, point, and coordinate levels. Extensive experiments show that FlashMesh achieves up to a 2 x speedup over standard autoregressive models while also improving generation fidelity. Our results demonstrate that structural priors in mesh data can be systematically harnessed to accelerate and enhance autoregressive generation.