GaussianPro is a novel method for 3D Gaussian Splatting (3DGS) that improves the densification of 3D Gaussians through a progressive propagation strategy. The method addresses the challenges of optimizing 3DGS in scenes with textureless areas, where traditional SfM techniques fail to provide sufficient initial points. GaussianPro leverages existing reconstructed geometries and patch matching to generate Gaussians with accurate positions and orientations. It operates in both 3D world space and 2D image space, rendering depth and normal maps to guide Gaussian densification. By iteratively propagating depth and normal values from neighboring pixels, GaussianPro identifies regions requiring additional Gaussians and initializes new ones based on these values. A planar loss is also incorporated to further regularize Gaussian orientations, enhancing the accuracy of the 3D geometry. Experiments on the Waymo and MipNeRF360 datasets show that GaussianPro significantly improves rendering quality, achieving a 1.15dB increase in PSNR compared to 3DGS. The method is robust to varying numbers of input images and produces more compact and accurate Gaussians, especially in textureless regions. GaussianPro outperforms 3DGS in both large-scale and small-scale scenes, demonstrating superior performance in rendering quality and efficiency. The method also shows improved results in indoor scenes with weak-texture surfaces. Overall, GaussianPro provides a more effective and efficient approach to 3D Gaussian Splatting, particularly in challenging scenarios with limited texture information.GaussianPro is a novel method for 3D Gaussian Splatting (3DGS) that improves the densification of 3D Gaussians through a progressive propagation strategy. The method addresses the challenges of optimizing 3DGS in scenes with textureless areas, where traditional SfM techniques fail to provide sufficient initial points. GaussianPro leverages existing reconstructed geometries and patch matching to generate Gaussians with accurate positions and orientations. It operates in both 3D world space and 2D image space, rendering depth and normal maps to guide Gaussian densification. By iteratively propagating depth and normal values from neighboring pixels, GaussianPro identifies regions requiring additional Gaussians and initializes new ones based on these values. A planar loss is also incorporated to further regularize Gaussian orientations, enhancing the accuracy of the 3D geometry. Experiments on the Waymo and MipNeRF360 datasets show that GaussianPro significantly improves rendering quality, achieving a 1.15dB increase in PSNR compared to 3DGS. The method is robust to varying numbers of input images and produces more compact and accurate Gaussians, especially in textureless regions. GaussianPro outperforms 3DGS in both large-scale and small-scale scenes, demonstrating superior performance in rendering quality and efficiency. The method also shows improved results in indoor scenes with weak-texture surfaces. Overall, GaussianPro provides a more effective and efficient approach to 3D Gaussian Splatting, particularly in challenging scenarios with limited texture information.