GeoCache: Training-Free Acceleration of Multi-View Texture Diffusion via Geometric Delta Transport

2026-08-13

AI Summary

This work addresses the high computational cost of multi-view texture diffusion models, which stems from repeated denoiser invocations, and the degradation in cross-view consistency often caused by existing training-free acceleration methods. The authors propose a plug-and-play, training-free acceleration approach that leverages cross-view geometric correspondences to transfer per-step denoising updates (deltas) from an anchor view to other views, effectively exploiting redundancy in signal evolution without modifying the model architecture or requiring retraining. By integrating periodic full-view correction with a sampler-consistent reconstruction strategy, the method achieves a 2.21× speedup on models such as Hunyuan3D-2.1 with the lowest FLOPs, while preserving high fidelity (MV-PSNR: 33.60 dB) and strong cross-view consistency (MV-LPIPS: 0.0293).

Abstract

Geometry-conditioned multi-view diffusion enables high-quality 3D texture generation, but its repeated per-view denoiser evaluations introduce substantial computational cost. Existing training-free accelerators primarily exploit temporal redundancy by reusing computation across denoising steps. In multi-view texturing, however, skipping a step also removes the cross-view interaction that continually aligns different observations of the same surface, leading to rapidly degraded consistency and fidelity. Our analysis identifies a complementary source of redundancy: although intermediate features remain view-specific, geometrically corresponding surface points exhibit transferable evolution in their predicted clean signals. Based on this observation, we introduce \gc{}, a training-free plugin that evaluates a rotating subset of anchor views and transports their geometry-aligned per-step $\xz$ updates to the remaining views. Periodic full-view computation controls accumulated error, while sampler-consistent reconstruction preserves the denoising trajectory. \gc{} requires neither retraining nor architectural modification and uses the position maps already available in geometry-conditioned texturing pipelines. Across Hunyuan3D-2.1, SyncMVD, and MVPainter, \gc{} achieves a stronger speed--fidelity trade-off than temporal caches and step reduction at operating points above $2\times$. On Hunyuan3D-2.1, it delivers a $2.21\times$ denoiser-loop speedup with an MV-LPIPS of 0.0293 and an MV-PSNR of 33.60 dB, providing the best fidelity among all tested methods above $2\times$. The same transferred configuration reaches the highest speedup and lowest FLOPs on SyncMVD, while \gc{} achieves the lowest FLOPs and best fidelity among the accelerated methods on MVPainter. These results establish cross-view geometry as an effective acceleration axis for multi-view texture diffusion.

Problem

• multi-view diffusion
• 3D texture generation
• computational acceleration
• cross-view consistency
• geometry-conditioned generation

Innovation

• geometric delta transport
• training-free acceleration
• multi-view diffusion
• cross-view consistency
• texture generation