ODGS: 3D Scene Reconstruction from Omnidirectional Images with 3D Gaussian Splattings

TL;DR

This work presents ODGS, a novel rasterization pipeline for omnidirectional images, with geometric interpretation that reveals the implicit assumptions within the proposed pipeline, which is verified through mathematical proofs.

Abstract

Omnidirectional (or 360-degree) images are increasingly being used for 3D applications since they allow the rendering of an entire scene with a single image. Existing works based on neural radiance fields demonstrate successful 3D reconstruction quality on egocentric videos, yet they suffer from long training and rendering times. Recently, 3D Gaussian splatting has gained attention for its fast optimization and real-time rendering. However, directly using a perspective rasterizer to omnidirectional images results in severe distortion due to the different optical properties between two image domains. In this work, we present ODGS, a novel rasterization pipeline for omnidirectional images, with geometric interpretation. For each Gaussian, we define a tangent plane that touches the unit sphere and is perpendicular to the ray headed toward the Gaussian center. We then leverage a perspective camera rasterizer to project the Gaussian onto the corresponding tangent plane. The projected Gaussians are transformed and combined into the omnidirectional image, finalizing the omnidirectional rasterization process. This interpretation reveals the implicit assumptions within the proposed pipeline, which we verify through mathematical proofs. The entire rasterization process is parallelized using CUDA, achieving optimization and rendering speeds 100 times faster than NeRF-based methods. Our comprehensive experiments highlight the superiority of ODGS by delivering the best reconstruction and perceptual quality across various datasets. Additionally, results on roaming datasets demonstrate that ODGS restores fine details effectively, even when reconstructing large 3D scenes. The source code is available on our project page (https://github.com/esw0116/ODGS).

Figures (5)

• Figure 1: Illustration on rasterization process of ODGS. We describe the process of projecting a 3D Gaussian to the omnidirectional pixel space. (a) The coordinate is transformed from the original camera pose (black) to the target Gaussian (green), making the $z$-axis of the coordinate heads towards the center of the Gaussian. (b) The Gaussian is projected onto the corresponding tangent plane. (c) The projected Gaussian is horizontally stretched when transformed into equirectangular space. (d) The Gaussian in equirectangular space is linearly transformed to the pixel space, followed by combination with the other projected Gaussian.
• Figure 2: Changes of PSNR, SSIM, and LPIPS over optimization time for each method. ODGS shows the best result as well as the highest convergence speed in both scenes.
• Figure 3: Qualitative comparisons in the egocentric scenes (10 min.). Each scene is brought from Ricoh360, OmniBlender, and OmniPhotos, respectively. Best viewed when zoomed in.
• Figure 4: Qualitative comparisons in the roaming scenes (10 min.). Each scene is brought from 360Roam, OmniScenes, and 360VO, respectively. Best viewed when zoomed in.
• Figure 5: Qualitative comparison of rendered images according to the Gaussian densification policy during optimization.