Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration

This paper proposes Fast3Dcache, a training-free geometry-aware caching framework that accelerates 3D diffusion inference while preserving geometric fidelity. The method introduces a Predictive Caching Scheduler Constraint (PCSC) to dynamically determine cache quotas and a Spatiotemporal Stability Criterion (SSC) to select stable features for reuse. Comprehensive experiments show that Fast3Dcache achieves up to a 27.12% speed-up and a 54.8% reduction in FLOPs, with minimal degradation in geometric quality.

PDF | Project Page

Recommended citation: M Yang, Y Yang, C Xu, C Song, Y Zuo, T Zhao, R Li, C Zhang. (2025). "Fast3Dcache: Training-free 3D Geometry Synthesis Acceleration." arXiv preprint arXiv:2511.22533.
Download Paper