[论文] StructSplat: Generalizable 3D Gaussian Splatting from Uncalibrated Spa...

论文概要

研究领域: CV 作者: Jia-Chen Zhao, Beiqi Chen, Xinyang Chen 发布时间: 2026-06-26 arXiv: 2606.28321

中文摘要

我们提出 StructSplat，一种前馈且可泛化的3D高斯重建框架，直接在未标定图像上运行，无需相机参数。现有方法要么依赖逐场景优化，要么假设已知相机位姿，并经常将几何和外观纠缠在统一的主干中，限制了重建保真度和泛化能力。我们的关键思想是采用结构化表示，在重建过程中为几何、语义和纹理线索组织显式角色。具体来说，我们引入像素对齐的特征注入机制以实现从2D观察的准确纹理建模，结合语义感知先验以改善全局一致性，并设计相机对齐策略以防止信息泄漏并提高泛化能力。实验表明，我们的方法在挑战性基准上显著优于先前方法。在DL3DV上，我们的方法达到28.045 PSNR，超过AnySplat（22.377）+5.67 dB。在跨数据集评估中，我们的方法在ACID上比AnySplat高+1.94 dB，在RealEstate10K上高+1.72 dB。

原文摘要

We present StructSplat, a feed-forward and generalizable 3D Gaussian reconstruction framework that operates directly on uncalibrated images without requiring camera parameters. Existing methods either rely on per-scene optimization or assume known camera poses, and often entangle geometry and appearance within a unified backbone, limiting reconstruction fidelity and generalization. Our key idea is to adopt a structured representation that organizes geometry, semantic, and texture cues with explicit roles in the reconstruction process. Specifically, we introduce a pixel-aligned feature injection mechanism to enable accurate texture modeling from 2D observations, incorporate semantic-aware priors to improve global consistency, and design a camera alignment strategy to prevent information lea...

--- *自动采集于 2026-06-30*

#论文 #arXiv #CV #小凯