RAIN-GS is a novel optimization strategy for 3D Gaussian splatting (3DGS) that relaxes the requirement for accurate initialization. The method is based on an in-depth analysis of the original 3DGS optimization scheme and SfM initialization in the frequency domain. RAIN-GS consists of three key components: 1) sparse-large-variance (SLV) initialization, 2) progressive Gaussian low-pass filtering, and 3) adaptive bound-expanding split (ABE-Split) algorithm. SLV initialization starts with a sparse set of Gaussians with large variance, enabling the learning of low-frequency components. Progressive Gaussian low-pass filtering ensures that Gaussians cover larger areas during early training, while ABE-Split encourages Gaussians to move further from their initial positions. RAIN-GS effectively relaxes the need for accurate initialization, achieving performance on-par or better than 3DGS trained with accurate SfM point clouds. The method is evaluated on multiple datasets, demonstrating its effectiveness in real-world scenarios where initial point clouds may be noisy or unavailable. RAIN-GS opens up new possibilities for applying 3DGS in challenging settings by reducing the reliance on accurate initial point clouds.RAIN-GS is a novel optimization strategy for 3D Gaussian splatting (3DGS) that relaxes the requirement for accurate initialization. The method is based on an in-depth analysis of the original 3DGS optimization scheme and SfM initialization in the frequency domain. RAIN-GS consists of three key components: 1) sparse-large-variance (SLV) initialization, 2) progressive Gaussian low-pass filtering, and 3) adaptive bound-expanding split (ABE-Split) algorithm. SLV initialization starts with a sparse set of Gaussians with large variance, enabling the learning of low-frequency components. Progressive Gaussian low-pass filtering ensures that Gaussians cover larger areas during early training, while ABE-Split encourages Gaussians to move further from their initial positions. RAIN-GS effectively relaxes the need for accurate initialization, achieving performance on-par or better than 3DGS trained with accurate SfM point clouds. The method is evaluated on multiple datasets, demonstrating its effectiveness in real-world scenarios where initial point clouds may be noisy or unavailable. RAIN-GS opens up new possibilities for applying 3DGS in challenging settings by reducing the reliance on accurate initial point clouds.