The paper introduces 360-GS, a novel method for panoramic Gaussian splatting that enables real-time rendering of indoor scenes from panoramic images. Traditional 3D Gaussian Splatting (3D-GS) struggles with panoramic inputs due to the challenges of projecting 3D Gaussians onto a spherical surface and handling sparse input data. 360-GS addresses these issues by projecting 3D Gaussians onto the tangent plane of the unit sphere and then mapping them to the spherical surface, avoiding the need for complex projections. The method leverages room layout priors, which are rich in structural information, to guide the initialization and regularization of 3D Gaussians, improving the quality of novel view synthesis. Experimental results demonstrate that 360-GS outperforms state-of-the-art methods in terms of visual quality and robustness to sparse input data, providing an immersive experience for indoor roaming applications.The paper introduces 360-GS, a novel method for panoramic Gaussian splatting that enables real-time rendering of indoor scenes from panoramic images. Traditional 3D Gaussian Splatting (3D-GS) struggles with panoramic inputs due to the challenges of projecting 3D Gaussians onto a spherical surface and handling sparse input data. 360-GS addresses these issues by projecting 3D Gaussians onto the tangent plane of the unit sphere and then mapping them to the spherical surface, avoiding the need for complex projections. The method leverages room layout priors, which are rich in structural information, to guide the initialization and regularization of 3D Gaussians, improving the quality of novel view synthesis. Experimental results demonstrate that 360-GS outperforms state-of-the-art methods in terms of visual quality and robustness to sparse input data, providing an immersive experience for indoor roaming applications.