ErpGS: Equirectangular Image Rendering Enhanced with 3D Gaussian Regularization

Shintaro Ito1, Natsuki Takama1, Koichi Ito1, Hwann-Tzong Chen2, Takafumi Aoki1
1Tohoku University, 2National Tsing Hua University
ICIP 2025
Paper Code (Coming Soon) arXiv

Abstract

The use of multi-view images acquired by a 360-degree camera can reconstruct a 3D space with a wide area. There are 3D reconstruction methods from equirectangular images based on NeRF and 3DGS, as well as Novel View Synthesis (NVS) methods. On the other hand, it is necessary to overcome the large distortion caused by the projection model of a 360-degree camera when equirectangular images are used. In 3DGS-based methods, the large distortion of the 360-degree camera model generates extremely large 3D Gaussians, resulting in poor rendering accuracy. We propose ErpGS, which is Omnidirectional GS based on 3DGS to realize NVS addressing the problems. ErpGS introduce some rendering accuracy improvement techniques: geometric regularization, scale regularization, and distortion-aware weights and a mask to suppress the effects of obstacles in equirectangular images. Through experiments on public datasets, we demonstrate that ErpGS can render novel view images more accurately than conventional methods.

Overview of ErpGS

Overview of ErpGS

We propose ErpGS, which is Omnidirectional GS based on 3DGS to realize NVS addressing the problems.

Quantitative Results on NVS

Quantitative results of RGB rendering at novel viewpoints compared with EgoNeRF , ODGS , OmniGS and proposed method (Ours). In this table, LPIPS is based on AlexNet to encode rendered images.

Quantitative results

Qualitative Results on NVS

Qualitative Results

Experimental results of rendered ERP images at novel viewpoints on several datasets.

Ablation Studies

Ablation studies for proposed components in terms of the quality of rendered RGB on OmniBlender.

Ablation Studies Quantitative Results

Ablation Studies

Qualitative results of ablation studies: (a) Rendered normal maps, (b) Gaussian ellipsoids and rendered RGB. ‘Overview’ means 3D Gaussians seen from a distance in a target scene. Blue points depict centers of each 3D Gaussian. ‘Rings’ means the visualized 3D Gaussians with rings. In the result of w/o Ls, rendered quality degraded due to large 3D Gaussians.

BibTeX


        @article{Ito-arXiv-2025,
          author = "Ito, S. and Takama, N. and Ito, K. and Chen, H.-T. and Aoki, T.",
          title  = "ErpGS: Equirectangular Image Rendering Enhanced with 3D Gaussian Regularization",
          journal = CoRR,
          volume = "abs/2505.19883",
          month = may,
          year = "2025"
        }