GES: Generalized Exponential Splatting for Efficient Radiance Field Rendering

Abdullah Hamdi,Luke Melas-Kyriazi,Jinjie Mai,Guocheng Qian,Ruoshi Liu,Carl Vondrick,Bernard Ghanem,Andrea Vedaldi

2024-05-25

Abstract:Advancements in 3D Gaussian Splatting have significantly accelerated 3D reconstruction and generation. However, it may require a large number of Gaussians, which creates a substantial memory footprint. This paper introduces GES (Generalized Exponential Splatting), a novel representation that employs Generalized Exponential Function (GEF) to model 3D scenes, requiring far fewer particles to represent a scene and thus significantly outperforming Gaussian Splatting methods in efficiency with a plug-and-play replacement ability for Gaussian-based utilities. GES is validated theoretically and empirically in both principled 1D setup and realistic 3D scenes.

Computer Vision and Pattern Recognition,Graphics,Machine Learning

What problem does this paper attempt to address?

The paper mainly proposes an improvement scheme for the problems existing in the 3D Gaussian Splatting method. Specifically, although 3D Gaussian Splatting has made significant progress in 3D reconstruction and generation, and has the advantage of a fast differentiable renderer, it requires a large number of Gaussians to represent complex 3D scenes, which leads to considerable memory usage. Moreover, since the Gaussian function is inherently a low-pass filter, it has limitations in representing signals with sharp edges, thus requiring more Gaussians to approximate these types of signals.

To address the aforementioned issues, the paper introduces Generalized Exponential Splatting (GES), a novel representation method that utilizes the Generalized Exponential Function (GEF) to model 3D scenes. Compared to traditional Gaussian functions, GEF allows for shape adjustment through an additional learning parameter β, enabling it to represent signals with sharp features, such as square or triangular signals, more effectively. This not only reduces the number of required particles but also improves efficiency and can serve as a plug-and-play replacement for Gaussian-based tools.

The main contributions of the paper are as follows:
1. **Theoretical Analysis**: The validity of using the Generalized Exponential Function instead of the Gaussian function for scene modeling is verified through numerical simulations.
2. **Introduction of GES**: A novel 3D representation method that utilizes GEF to develop a splatting-based approach suitable for real-time, efficient novel view synthesis.
3. **Experimental Validation**: A series of experiments demonstrate that compared to traditional Gaussian splatting, GES can reduce memory requirements by over 50% and increase rendering speed by up to 39%, while maintaining competitive performance.

In summary, the paper aims to address the inefficiency in representing complex signals and the excessive memory consumption issues in traditional 3D Gaussian Splatting methods by introducing GES.