The paper introduces GES (Generalized Exponential Splatting), a novel representation for 3D scene modeling that uses Generalized Exponential Functions (GEF) to model 3D scenes. GES aims to address the limitations of Gaussian Splatting, which requires a large number of Gaussians to represent scenes, leading to high memory usage. GES employs GEFs, which are more flexible and can model sharp edges and high-frequency signals more accurately. The method is validated both theoretically and empirically, showing superior performance in novel view synthesis benchmarks with reduced memory usage and increased rendering speed. The authors also propose a frequency-modulated image loss to help GES train from low-frequency to high-frequency details, achieving a 50% reduction in memory requirement and a 39% increase in rendering speed compared to Gaussian Splatting. The code for GES is available on the project website.The paper introduces GES (Generalized Exponential Splatting), a novel representation for 3D scene modeling that uses Generalized Exponential Functions (GEF) to model 3D scenes. GES aims to address the limitations of Gaussian Splatting, which requires a large number of Gaussians to represent scenes, leading to high memory usage. GES employs GEFs, which are more flexible and can model sharp edges and high-frequency signals more accurately. The method is validated both theoretically and empirically, showing superior performance in novel view synthesis benchmarks with reduced memory usage and increased rendering speed. The authors also propose a frequency-modulated image loss to help GES train from low-frequency to high-frequency details, achieving a 50% reduction in memory requirement and a 39% increase in rendering speed compared to Gaussian Splatting. The code for GES is available on the project website.