ConFixGS: Learning to Fix Feedforward 3D Gaussian Splatting with Confidence-Aware Diffusion Priors in Driving Scenes

2026-05-10

AI Summary

This work addresses the poor reconstruction quality of feed-forward 3D Gaussian Splatting (3DGS) in trajectory-sparse driving scenarios, where hallucinated or inconsistent content frequently arises. To this end, the authors propose a plug-and-play refinement method that, for the first time, integrates confidence-aware diffusion priors with multi-view consistency. Specifically, local pseudo-targets are generated using a diffusion model, and a dense confidence map is constructed via multi-view reprojection cross-validation. This confidence map guides a pre-trained feed-forward 3DGS model to enhance reliable geometric details while suppressing unreliable artifacts. Experiments on Waymo, nuScenes, and KITTI demonstrate significant improvements in novel view synthesis, achieving up to a 3.68 dB gain in PSNR and nearly a 50% reduction in FID, thereby overcoming limitations of existing approaches that rely on optimization or iterative refinement from nearby observed views.

Abstract

Feedforward 3D Gaussian Splatting (3DGS) often struggles in trajectory-based sparse-view driving scenes. Existing Gaussian repair methods mainly target optimization-based 3DGS, while diffusion-based repair is typically restricted to iterative refinement near observed viewpoints, leaving feedforward 3DGS repair underexplored. We propose ConFixGS, a plug-and-play method that learns to fix feedforward 3DGS with confidence-aware diffusion priors. Starting from a pretrained feedforward model, ConFixGS generates diffusion-enhanced local pseudo-targets and validates them through reprojection-based cross-checking against support views. The resulting dense confidence maps guide refinement, enhancing reliable details while suppressing hallucinated or inconsistent evidence. On Waymo, nuScenes, and KITTI, ConFixGS improves challenging novel view synthesis, with PSNR gains of up to 3.68 dB and FID reduced by nearly half. Our results highlight confidence-aware fusion of generative priors and support-view consistency as a key principle for robust feedforward 3D driving scene reconstruction.

Problem

• Feedforward 3D Gaussian Splatting
• Sparse-view driving scenes
• Diffusion priors
• Novel view synthesis
• 3D scene reconstruction

Innovation

• Feedforward 3D Gaussian Splatting
• Diffusion Priors
• Confidence-Aware Fusion
• Sparse-View Driving Scenes
• Novel View Synthesis