G²PS: Geometry and Gradient-based Partitioning for
Panoramic Outdoor Reconstruction

Abstract

Scaling 3D Gaussian Splatting (3DGS) to large outdoor scenes is costly in both data acquisition and computation. Adopting panoramic images with equirectangular projection (ERP) can reduce capture effort via their full 360° field of view, yet the resulting omnipresent visibility invalidates existing partitioning strategies that rely on local camera frustums, causing block-wise optimization to degenerate into global training. We propose PanoLOG, a two-stage coarse-to-fine framework equipped with a Geometry and Gradient-based Partitioning Strategy (G²PS) tailored for large-scale panoramic 3DGS reconstruction. PanoLOG combines panorama-specific geometric priors in a global coarse stage with an adaptive, importance-aware partitioning scheme in the refinement stage, enabling truly block-parallel training under 360° visibility. We further construct Pano360, the first large-scale panoramic outdoor benchmark for 3DGS reconstruction. Extensive experiments demonstrate that PanoLOG achieves state-of-the-art rendering quality while maintaining scalable, block-parallel training.

Full algorithmic details of G²PS will be disclosed in our forthcoming publication.

Visual Results

All videos above are rendered from our trained model.

Qualitative Comparisons

Quantitative Results

Comparisons on Public Datasets

Ablation Studies

Citation

@article{panolog2026,
  title   = {Geometry and Gradient-based Partitioning for Panoramic Outdoor Reconstruction},
  author  = {Chen, Weijian and Yao, Weibo and Zhang, Yuhang and Tang, Xiaolin and
             Wang, Guo and Zhang, Weijun and Gao, Xitong and Chen, Yihao and
             Qin, Hongde and Qi, Lu},
  journal = {arXiv preprint arXiv:2607.08769},
  year    = {2026}
}