StructSplat: Generalizable 3D Gaussian Splatting from Uncalibrated Sparse Views

2026-06-26

AI Summary

This work proposes a feed-forward 3D Gaussian splatting framework that achieves high-fidelity and generalizable 3D reconstruction from sparse, uncalibrated images without requiring camera calibration. The method decouples geometry, semantics, and texture through a structured representation and integrates pixel-aligned feature injection, semantic priors, and a camera pose alignment strategy designed to prevent information leakage. These components collectively enhance reconstruction accuracy and cross-dataset generalization. Notably, this is the first approach to achieve high-quality 3D Gaussian reconstruction under uncalibrated conditions, attaining a PSNR of 28.045 on DL3DV—outperforming AnySplat by 5.67 dB—and demonstrating superior cross-dataset performance with gains of 1.94 dB and 1.72 dB on ACID and RealEstate10K, respectively.

Abstract

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 leakage and improve generalization. Experiments show that our method significantly outperforms prior approaches on challenging benchmarks. On DL3DV, our method achieves 28.045 PSNR, surpassing AnySplat (22.377) by +5.67 dB. In cross-dataset evaluation, our method achieves +1.94 dB over AnySplat on ACID and +1.72 dB on RealEstate10K. Project page: https://structsplat.github.io Code: https://github.com/J-C-Zhao/StructSplat

Problem

• 3D Gaussian Splatting
• uncalibrated images
• sparse views
• generalizable reconstruction
• camera poses

Innovation

• 3D Gaussian Splatting
• uncalibrated sparse views
• structured representation
• pixel-aligned feature injection
• generalizable reconstruction