Textured-GS: Gaussian Splatting with Spatially Defined Color and Opacity

TL;DR

An innovative method for rendering Gaussian splatting that incorporates spatially defined color and opacity variations using Spherical Harmonics using Spherical Harmonics is introduced, promising more efficient and high-quality scene reconstructions.

Abstract

In this paper, we introduce Textured-GS, an innovative method for rendering Gaussian splatting that incorporates spatially defined color and opacity variations using Spherical Harmonics (SH). This approach enables each Gaussian to exhibit a richer representation by accommodating varying colors and opacities across its surface, significantly enhancing rendering quality compared to traditional methods. To demonstrate the merits of our approach, we have adapted the Mini-Splatting architecture to integrate textured Gaussians without increasing the number of Gaussians. Our experiments across multiple real-world datasets show that Textured-GS consistently outperforms both the baseline Mini-Splatting and standard 3DGS in terms of visual fidelity. The results highlight the potential of Textured-GS to advance Gaussian-based rendering technologies, promising more efficient and high-quality scene reconstructions. Our implementation is available at https://github.com/ZhentaoHuang/Textured-GS.

Figures (5)

• Figure 1: Comparison between view-dependent 3D Gaussian and our textured Gaussian: The original 3DGS kerbl20233d (a) assigns a single opacity value to each Gaussian and maintains a fixed color per view. In contrast, our Textured-GS (b) enables color variation across the Gaussian ellipsoidal surface. With the addition of an opacity channel (c), it can also represent non-ellipsoidal shapes.
• Figure 2: Illustration of the vectors used for SH calculation: The standard 3DGS employs the viewing vector $\bm{v}$ (blue), which remains constant for all pixels under a given view. In contrast, our proposed Textured-GS uses the parameterization vector $\bm{n}$ (orange), which varies across the Gaussian surface. Please note that $\bm{n}$ is normalized by the scale of the Gaussian of each dimension.
• Figure 3: Visual comparison of three real-world scenes. Areas for zoomed-in comparison are marked with red boxes on the left images, highlighting the missing details in 3DGS and the poor representation of sharp features in Mini-Splatting.
• Figure 4: The rendering quality comparison between our method and Mini-Splatting across various numbers of Gaussians. Our method consistently outperforms Mini-Splatting across all scales and three distinct metrics.
• Figure 5: Zoomed-in comparison for the Room scene from the Mip-NeRF 360 dataset: Our proposed method, which incorporates textured opacity and color, achieves superior results in rendering sharp features with the same number of Gaussians compared to Mini-Splatting. Additionally, it more effectively captures the background texture than both 3DGS and Mini-Splatting.