3D-HGS: 3D Half Gaussian Splatting

Haolin Li*, Jinyang Liu*, Mario Sznaier, Camps Octavia
Northeastern university
CVPR 2025
*Indicates Equal Contribution
Paper(coming soon) Supplementary(coming soon) Code arXiv
First Image
Second Image
Third Image

Illustration of the 3D-HG kernel, and the mapping of a pair of 3D Half-Gaussians to a 2D image.

Abstract

Photo-realistic image rendering from scene 3D reconstruction is a fundamental problem in 3D computer vision. This domain has seen considerable advancements owing to the advent of recent neural rendering techniques. These techniques predominantly aim to focus on learning volumetric representations of 3D scenes and refining these representations via loss functions derived from their rendering. Among these, 3D Gaussian Splatting (3D-GS) has emerged as a preferred method, surpassing Neural Radiance Fields' (NeRFs) quality and rendering speed. 3D-GS uses parameterized 3D Gaussians to model both spatial locations and color information, combined with a tile-based fast rendering technique. Despite its superior performance, using 3D Gaussian kernels has inherent limitations in accurately representing discontinuous functions, notably at edges and corners corresponding to shape discontinuities, and across varying textures due to color discontinuities. In this paper, we introduce 3D Half-Gaussian 3D-HGS kernels, which can be used as a plug-and-play kernel, to address this issue. Our experiments demonstrate their capability to improve the performance of current 3D-GS related methods and achieve state-of-the-art rendering quality performance on various datasets without compromising their rendering speed. The code and trained models will be available on GitHub.

Motivation

Image 1
Five Gaussians fitting as square
Image 2
Three HGs fitting as square
Image 3
Gaussian and Half-Gaussian in spatial domain
Image 4
Gaussian and Half-Gaussian in Frequency domain

Comparison of Half-Gaussian and Gaussian Kernels fitting a square function and their Fourier Transforms. (a): fit- ting a square function with 5 Gaussian kernels, and (b): fitting a square with 3 Half-Gaussian kernels. When approximating sharp edges, the Half-Gaussian kernels achieve a lower error loss (1.85) compared to Gaussian kernels. Figures (c) and (d) illustrate the Gaussian and Half-Gaussian kernels in both the spatial and fre- quency domains, where the Half-Gaussian demonstrates a higher bandwidth than the Gaussian kernel, indicating its superior ability to capture high-frequency components.

3D-GS
3D-HGS (ours)

Results

Quantitative comparison to the SOTA methods on real-world datasets
Results Graph
Performance (PSNR↑) versus rendering speed for sev- eral state-of-the-art methods with Gaussian ker- nels and the proposed half-Gaussian kernels on the Mip-NeRF360 dataset. In all cases, using half-Gaussian kernels resulted in significant PSNR improvements, with similar or better rendering speed than the corresponding 3D Gaussian-based method
Youtube embed code here -->

Poster(coming soon)

BibTeX

@article{li20243d,
  title={3d-hgs: 3d half-gaussian splatting},
  author={Li, Haolin and Liu, Jinyang and Sznaier, Mario and Camps, Octavia},
  journal={arXiv preprint arXiv:2406.02720},
  year={2024}
}