QuerySplat: Decoupling Geometry and Appearance Representations in 3DGS Prediction

2026-08-02

AI Summary

This work addresses the limitations of existing feed-forward 3D Gaussian Splatting (3DGS) methods, which often suffer from rendering blur and pose dependency due to entangled geometry-appearance modeling, structural redundancy, or spatial rigidity. To overcome these issues, we propose QuerySplat, the first query-based 3DGS framework that leverages a pre-trained vision foundation model to provide 3D priors. Our approach introduces a dual-branch query decoder that explicitly decouples geometry and appearance modeling, enabling pose-independent feed-forward rendering. Evaluated on the DL3DV benchmark, QuerySplat achieves state-of-the-art performance, improving average PSNR by 2.30 dB over the best pose-free baseline and by 1.04 dB over pose-conditioned methods, while significantly enhancing detail recovery and rendering sharpness.

Abstract

While feed-forward 3D Gaussian Splatting (3DGS) enables efficient 3D reconstruction, achieving high-fidelity rendering remains challenging. Existing pixel-aligned approaches suffer from spatial inflexibility and massive structural redundancy, whereas query-based methods lack 3D priors and entangle geometry with appearance, yielding blurry, pose-dependent results. To overcome these deficiencies, we propose \textbf{QuerySplat}, a feed-forward 3DGS framework driven by geometric priors and explicit appearance decoupling. Specifically, we design a dual-branch query-based decoder: the geometry branch leverages a pretrained Vision Geometric Model for spatial understanding, which intrinsically endows QuerySplat with pose-free modeling capabilities, while the appearance branch recovers high-frequency details through a dedicated pathway separated from geometric attribute regression. Extensive experiments demonstrate that QuerySplat mitigates the blurry rendering issues of early query-based models and consistently outperforms pixel-aligned approaches in rendering fidelity. On the challenging DL3DV benchmark, it achieves state-of-the-art novel view synthesis performance, with average PSNR gains of 2.30 dB and 1.04 dB over the best pose-free and pose-required baselines, respectively. Project Page: https://inspatio.github.io/querysplat.

Problem

• 3D Gaussian Splatting
• geometry-appearance decoupling
• query-based reconstruction
• novel view synthesis
• pose-dependent rendering

Innovation

• 3D Gaussian Splatting
• geometry-appearance decoupling
• query-based reconstruction
• pose-free modeling
• novel view synthesis