CityGaussianV2: Efficient and Geometrically Accurate Reconstruction for Large-Scale Scenes
Yang Liu, Chuanchen Luo, Zhongkai Mao, Junran Peng, Zhaoxiang Zhang
2024-11-04
Summary
This paper presents CityGaussianV2, a new method for efficiently and accurately reconstructing large-scale 3D scenes. It improves upon existing techniques by addressing challenges related to image quality and computational efficiency.
What's the problem?
While recent advancements in 3D Gaussian Splatting (3DGS) have made it easier to create realistic images from 3D data, accurately representing complex surfaces in large scenes is still difficult. The unstructured nature of 3DGS can lead to blurry images and slow processing times, making it hard to use in real-world applications.
What's the solution?
CityGaussianV2 introduces several innovative techniques to enhance the reconstruction process. It uses a method called decomposed-gradient-based densification to reduce blurriness and improve image clarity. Additionally, it implements an elongation filter to manage the number of Gaussian points used in the process, preventing overload during training. The pipeline is optimized for parallel training, which significantly speeds up the process and reduces memory usage while maintaining high visual quality.
Why it matters?
This research is important because it provides a more effective way to reconstruct large-scale scenes, which can be applied in various fields like gaming, virtual reality, and urban planning. By improving both the quality of the images and the efficiency of the reconstruction process, CityGaussianV2 enables better visualization of complex environments, making it a valuable tool for developers and researchers.
Abstract
Recently, 3D Gaussian Splatting (3DGS) has revolutionized radiance field reconstruction, manifesting efficient and high-fidelity novel view synthesis. However, accurately representing surfaces, especially in large and complex scenarios, remains a significant challenge due to the unstructured nature of 3DGS. In this paper, we present CityGaussianV2, a novel approach for large-scale scene reconstruction that addresses critical challenges related to geometric accuracy and efficiency. Building on the favorable generalization capabilities of 2D Gaussian Splatting (2DGS), we address its convergence and scalability issues. Specifically, we implement a decomposed-gradient-based densification and depth regression technique to eliminate blurry artifacts and accelerate convergence. To scale up, we introduce an elongation filter that mitigates Gaussian count explosion caused by 2DGS degeneration. Furthermore, we optimize the CityGaussian pipeline for parallel training, achieving up to 10times compression, at least 25% savings in training time, and a 50% decrease in memory usage. We also established standard geometry benchmarks under large-scale scenes. Experimental results demonstrate that our method strikes a promising balance between visual quality, geometric accuracy, as well as storage and training costs. The project page is available at https://dekuliutesla.github.io/CityGaussianV2/.