360-GeoGS: Geometrically Consistent Feed-Forward 3D Gaussian Splatting Reconstruction for 360 Images

TL;DR

This paper introduces a feed-forward 3D Gaussian Splatting method for 360 images that achieves both high-quality rendering and improved geometric consistency, enhancing 3D reconstruction accuracy.

Contribution

It proposes a novel framework with depth-normal regularization to ensure geometric consistency in 3D Gaussian Splatting for 360 images, without sacrificing rendering quality.

Findings

Significantly improves geometric consistency in 3D reconstructions.

Maintains high rendering quality comparable to state-of-the-art methods.

Effective for spatial perception applications like AR and robotics.

Abstract

3D scene reconstruction is fundamental for spatial intelligence applications such as AR, robotics, and digital twins. Traditional multi-view stereo struggles with sparse viewpoints or low-texture regions, while neural rendering approaches, though capable of producing high-quality results, require per-scene optimization and lack real-time efficiency. Explicit 3D Gaussian Splatting (3DGS) enables efficient rendering, but most feed-forward variants focus on visual quality rather than geometric consistency, limiting accurate surface reconstruction and overall reliability in spatial perception tasks. This paper presents a novel feed-forward 3DGS framework for 360 images, capable of generating geometrically consistent Gaussian primitives while maintaining high rendering quality. A Depth-Normal geometric regularization is introduced to couple rendered depth gradients with normal information, supervising Gaussian rotation, scale, and position to improve point cloud and surface accuracy. Experimental results show that the proposed method maintains high rendering quality while significantly improving geometric consistency, providing an effective solution for 3D reconstruction in spatial perception tasks.