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

2026-03-26

AI Summary

This work addresses the challenge of poor reconstruction quality in pose-free novel view synthesis (NVS) when directly applying existing 3D Vision Foundation Models (3DVFMs), which often suffer from geometric and pose inconsistencies. To this end, we propose AirSplat, the first framework that effectively transfers geometric priors from 3DVFMs to high-quality NVS without requiring input camera poses. Built upon 3D Gaussian splatting, AirSplat introduces a Self-Consistent Pose Alignment (SCPA) mechanism to resolve pose-geometry misalignment and designs a Score-based Opacity Matching (ROM) strategy that leverages geometry consistency scores guided by a sparse-view teacher model to filter out low-quality Gaussians. Evaluated on large-scale benchmarks, AirSplat significantly outperforms existing pose-free NVS methods, achieving state-of-the-art performance in both reconstruction accuracy and rendering fidelity.

Technology Category

• Computer Vision: 3D Computer Vision
• Humans and AI: Human-Aware Planning and Behavior Prediction
• Planning, Routing, and Scheduling: Replanning and Plan Repair

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

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.

Problem

• 3D Vision Foundation Models
• novel view synthesis
• pose-free
• 3D Gaussian Splatting
• visual geometry estimation

Innovation

• 3D Gaussian Splatting
• Pose-Free Novel View Synthesis
• Self-Consistent Pose Alignment
• Rating-based Opacity Matching
• 3D Vision Foundation Models