AirSplat: Alignment and Rating for Robust Feed-Forward 3D Gaussian Splatting

TL;DR

AirSplat is a novel training framework that enhances zero-shot 3D view synthesis by integrating geometric priors from foundation models, using pose alignment and opacity filtering to improve reconstruction quality.

Contribution

It introduces Self-Consistent Pose Alignment and Rating-based Opacity Matching to adapt 3DVFMs for high-fidelity, pose-free novel view synthesis.

Findings

Outperforms state-of-the-art pose-free NVS methods in benchmarks

Effectively filters degraded primitives using local geometry knowledge

Enables simultaneous geometry estimation and view synthesis

Abstract

While 3D Vision Foundation Models (3DVFMs) have demonstrated remarkable zero-shot capabilities in visual geometry estimation, their direct application to generalizable novel view synthesis (NVS) remains challenging. In this paper, we propose AirSplat, a novel training framework that effectively adapts the robust geometric priors of 3DVFMs into high-fidelity, pose-free NVS. Our approach introduces two key technical contributions: (1) Self-Consistent Pose Alignment (SCPA), a training-time feedback loop that ensures pixel-aligned supervision to resolve pose-geometry discrepancy; and (2) Rating-based Opacity Matching (ROM), which leverages the local 3D geometry consistency knowledge from a sparse-view NVS teacher model to filter out degraded primitives. Experimental results on large-scale benchmarks demonstrate that our method significantly outperforms state-of-the-art pose-free NVS approaches in reconstruction quality. Our AirSplat highlights the potential of adapting 3DVFMs to enable simultaneous visual geometry estimation and high-quality view synthesis.