The paper introduces GaussianPro, a novel method for 3D Gaussian Splatting (3DGS) that addresses the challenges of optimizing 3D Gaussians, particularly in textureless areas. Traditional 3DGS relies heavily on sparse point clouds from Structure-from-Motion (SfM) techniques, which often fail to provide sufficient points in textureless regions, leading to poor initialization and low-quality renderings. To overcome this, GaussianPro employs a progressive propagation strategy that leverages existing reconstructed geometries and patch matching techniques to guide the densification of 3D Gaussians. This approach ensures more accurate and compact Gaussians, especially in low-texture regions, by transferring geometric information from well-modeled areas to under-modeled areas. The method is validated on large-scale and small-scale scenes, demonstrating significant improvements over 3DGS in terms of PSNR, with a notable 1.15 dB gain on the Waymo dataset. The contributions of GaussianPro include a novel Gaussian propagation strategy, a planar loss for geometric regularization, and robust performance across varying numbers of input images.The paper introduces GaussianPro, a novel method for 3D Gaussian Splatting (3DGS) that addresses the challenges of optimizing 3D Gaussians, particularly in textureless areas. Traditional 3DGS relies heavily on sparse point clouds from Structure-from-Motion (SfM) techniques, which often fail to provide sufficient points in textureless regions, leading to poor initialization and low-quality renderings. To overcome this, GaussianPro employs a progressive propagation strategy that leverages existing reconstructed geometries and patch matching techniques to guide the densification of 3D Gaussians. This approach ensures more accurate and compact Gaussians, especially in low-texture regions, by transferring geometric information from well-modeled areas to under-modeled areas. The method is validated on large-scale and small-scale scenes, demonstrating significant improvements over 3DGS in terms of PSNR, with a notable 1.15 dB gain on the Waymo dataset. The contributions of GaussianPro include a novel Gaussian propagation strategy, a planar loss for geometric regularization, and robust performance across varying numbers of input images.