SyncFix: Fixing 3D Reconstructions via Multi-View Synchronization

TL;DR

SyncFix is a novel framework that enhances 3D scene reconstructions by enforcing multi-view consistency during diffusion-based refinement, improving quality and robustness without requiring clean references.

Contribution

It introduces a joint latent bridge matching approach for multi-view synchronization, generalizes to arbitrary views, and surpasses state-of-the-art methods in reconstruction quality.

Findings

Reconstruction quality improves with more views, with diminishing returns.

SyncFix outperforms current state-of-the-art baselines.

High-fidelity reconstructions achieved even without clean references.

Abstract

We present SyncFix, a framework that enforces cross-view consistency during the diffusion-based refinement of reconstructed scenes. SyncFix formulates refinement as a joint latent bridge matching problem, synchronizing distorted and clean representations across multiple views to fix the semantic and geometric inconsistencies. This means SyncFix learns a joint conditional over multiple views to enforce consistency throughout the denoising trajectory. Our training is done only on image pairs, but it generalizes naturally to an arbitrary number of views during inference. Moreover, reconstruction quality improves with additional views, with diminishing returns at higher view counts. Qualitative and quantitative results demonstrate that SyncFix consistently generates high-quality reconstructions and surpasses current state-of-the-art baselines, even in the absence of clean reference images. SyncFix achieves even higher fidelity when sparse references are available.