GaussianPro: 3D Gaussian Splatting with Progressive Propagation

1University of Science and Technology of China, 2The University of Hong Kong, 3Nanjing University, 4The University of Adelaide, 5ShanghaiTech University 6Texas A&M University,

                                                3DGS (left)                                                                            GaussianPro (right)


The advent of 3D Gaussian Splatting (3DGS) has recently brought about a revolution in the field of neural rendering, facilitating high-quality renderings at real-time speed. However, 3DGS heavily depends on the initialized point cloud produced by Structure-from-Motion (SfM) techniques. When tackling with large-scale scenes that unavoidably contain texture-less surfaces, the SfM techniques always fail to produce enough points in these surfaces and cannot provide good initialization for 3DGS. As a result, 3DGS suffers from difficult optimization and low-quality renderings. In this paper, inspired by classical multi-view stereo (MVS) techniques, we propose GaussianPro, a novel method that applies a progressive propagation strategy to guide the densification of the 3D Gaussians. Compared to the simple split and clone strategies used in 3DGS, our method leverages the priors of the existing reconstructed geometries of the scene and patch matching techniques to produce new Gaussians with accurate positions and orientations.

Experiments on both large-scale and small-scale scenes validate the effectiveness of our method, where our method significantly surpasses 3DGS on the Waymo dataset, exhibiting an improvement of 1.15dB in terms of PSNR.


The sparse SfM points and less-constrained densification strategies of 3DGS pose challenges in optimizing 3D Gaussians, particularly for textureless areas. 3DGS generates incorrect Gaussians (blue circle) to be over-fitted on the training images, leading to a noticeable performance drop in novel view rendering with erroneous geometries.


More Results

Note that no prior used in our GaussianPro.

                    Rendered Images                                          Rendered Depth                                           Rendered Normal

Gaussian Visualization

Please note that we did not perform any additional processing on the sky, but it is possible to separate the sky from the foreground by segmenting them as an independent layer during modeling.

                                Ours GaussianPro (left)                                                                            3DGS (right)


  title={GaussianPro: 3D Gaussian Splatting with Progressive Propagation},
  author={Cheng, Kai and Long, Xiaoxiao and Yang, Kaizhi and Yao, Yao and Yin, Wei and Ma, Yuexin and Wang, Wenping and Chen, Xuejin},
  journal={arXiv preprint arXiv:2402.14650},