EDGS: Eliminating Densification for Efficient Convergence of 3DGS

TL;DR

EDGS tackles slow convergence in 3D Gaussian Splatting caused by iterative densification by replacing it with a dense initialization derived from triangulated dense 2D correspondences across views. The method places Gaussians directly in 3D and assigns informed position, color, and scale from the start, reducing the optimization path and removing the need for densification. This approach yields faster convergence and higher rendering quality while using roughly half the final splats, and is compatible with other 3DGS acceleration techniques. Across Mip-NeRF360, Tanks & Temples, and Deep Blending, EDGS outperforms both speed-optimized and quality-focused baselines, enabling high-fidelity reconstructions with significantly reduced training time.

Abstract

3D Gaussian Splatting reconstructs scenes by starting from a sparse Structure-from-Motion initialization and iteratively refining under-reconstructed regions. This process is inherently slow, as it requires multiple densification steps where Gaussians are repeatedly split and adjusted, following a lengthy optimization path. Moreover, this incremental approach often leads to suboptimal renderings, particularly in high-frequency regions where detail is critical. We propose a fundamentally different approach: we eliminate densification process with a one-step approximation of scene geometry using triangulated pixels from dense image correspondences. This dense initialization allows us to estimate rough geometry of the scene while preserving rich details from input RGB images, providing each Gaussian with well-informed colors, scales, and positions. As a result, we dramatically shorten the optimization path and remove the need for densification. Unlike traditional methods that rely on sparse keypoints, our dense initialization ensures uniform detail across the scene, even in high-frequency regions where 3DGS and other methods struggle. Moreover, since all splats are initialized in parallel at the start of optimization, we eliminate the need to wait for densification to adjust new Gaussians. Our method not only outperforms speed-optimized models in training efficiency but also achieves higher rendering quality than state-of-the-art approaches, all while using only half the splats of standard 3DGS. It is fully compatible with other 3DGS acceleration techniques, making it a versatile and efficient solution that can be integrated with existing approaches.

Figures (14)

• Figure 1: Visual comparison of initialization methods on the stump scene from the Mip-NeRF360 dataset barron_mip-nerf_2022. The left image represents ground truth. The middle image shows the traditional 3DGS approach initialization with Structure-from-Motion (SfM) schonberger2016structure. The right image illustrates initialization with our method using matchings. Despite noisy appearance at the initialization, our model can jointly optimize all the gaussians and achieve better reconstruction quality.
• Figure 2: Qualitative comparison on flowers and treehill from Mip-NeRF360 barron_mip-nerf_2022, train from Tank & Temples knapitsch2017tanks and Playroom from Deep Blending hedman2018deep. For this visualization, we crop regions of interest. See supplementary materials for full renderings. Our model effectively reduces blur and preserves fine details that other methods often overlook or blur. It also performs comparably to or better than state-of-the-art methods, achieving faster convergence. For comparison, we additionally provide renderings of these cropped regions for our model with 3DGS trained for only 3,000 steps.
• Figure 3: Extreme viewpoint rendering. EDGS (right) better preserves details and reduces stretched Gaussians when rendering from viewpoints far outside the training set compared to the 3DGS (left). This results in a more consistent distribution and improved quality, especially in challenging regions like the building and flower pot.
• Figure 4: Distributions of 3DGS parameters change in color/coordinate space throughout training. We compare color $\bm{g}_i^c$ and coordinate $\bm{g}_i^x$ changes from initialization to 30K steps. Our method not only initializes closer to the solution (right chart) but also requires significantly fewer adjustments (left chart) through the ptimisation process, leading to faster and more stable convergence.
• Figure 5: The effect of adding noise $\mathcal{N}(0,\sigma)$ to our model. Please note that noise scale $\sigma$ is higher for color noise. In the supplementary, we provide images visualizing the effect of noise on the initialization quality.