EGGS: Edge Guided Gaussian Splatting for Radiance Fields

TL;DR

EGGS addresses the lack of edge awareness in Gaussian splatting losses by introducing a gradient-based edge weight $\\phi(u,v)=1+\\beta\\|\\nabla im(u,v)\\|_p$ and a corresponding loss $Loss(c,im)=(1-\\lambda)\\|\\phi(c-im)\\|_1+\\lambda D_{SSIM}(c,im)$. This edge-guided, image-space weighting biases particles toward edges without adding computation, and can be plugged into existing 3D Gaussian splatting frameworks. Empirical results across Banana, Train, and Truck datasets show PSNR improvements of about $2.1$, $1.2$, and $1.1$ dB, respectively, with qualitative gains such as sharper edges and edge-aligned particle placement. The approach is simple, widely applicable, and improves geometry and rendering fidelity in radiance-field representations while preserving the efficiency advantages of Gaussian splatting.

Abstract

The Gaussian splatting methods are getting popular. However, their loss function only contains the $\ell_1$ norm and the structural similarity between the rendered and input images, without considering the edges in these images. It is well-known that the edges in an image provide important information. Therefore, in this paper, we propose an Edge Guided Gaussian Splatting (EGGS) method that leverages the edges in the input images. More specifically, we give the edge region a higher weight than the flat region. With such edge guidance, the resulting Gaussian particles focus more on the edges instead of the flat regions. Moreover, such edge guidance does not crease the computation cost during the training and rendering stage. The experiments confirm that such simple edge-weighted loss function indeed improves about $1\sim2$ dB on several difference data sets. With simply plugging in the edge guidance, the proposed method can improve all Gaussian splatting methods in different scenarios, such as human head modeling, building 3D reconstruction, etc.

Figures (5)

• Figure 1: The edge guide in each view image can force more particles on the edge region, improving the scene accuracy.
• Figure 2: The PSNR during the training process (the lines are smoothed for better visualization). The blue line is 3DGS and the red line is EGGS. The EGGS can achieve better PSNR.
• Figure 3: The detailed difference between 3DGS and EGGS on the banana data set. The top row shows that the edges in the EGGS are much clearer than the edges in 3DGS. The bottom row shows that EGGS puts more particles near the edges.
• Figure 4: The edges guidance for the train and truck data set.
• Figure 5: The PSNR during the training process (the lines are smoothed for better visualization). The blue line is 3DGS and the red line is EGGS. The proposed EGGS can achieve better results on the train data.