The paper "Reducing the Memory Footprint of 3D Gaussian Splatting" by Panagiotis Papantonakis, Georgios Kopanas, Bernhard Kerbl, Alexandre Lanvin, and George Drettakis addresses the high memory requirements of 3D Gaussian Splatting (3DGS) for novel view synthesis (NVS). The authors identify three main areas contributing to the high memory usage: the number of 3D Gaussian primitives, the number of spherical harmonic (SH) coefficients, and the precision needed for storing Gaussian primitive attributes. They propose a comprehensive solution to reduce these memory requirements: 1. **Primitive Pruning**: An efficient, resolution-aware primitive pruning approach that reduces the primitive count by half. 2. **Adaptive SH Adjustment**: A method to choose the appropriate number of SH bands for each primitive, significantly reducing the overall memory footprint. 3. **Codebook-Based Quantization**: A codebook quantization method and the use of half-float representation to further reduce the stored size of the representation. These three components result in a 27× reduction in overall disk size and a 1.7× increase in rendering speed. The authors demonstrate their method on standard datasets and show that it significantly reduces download times on mobile devices, achieving a 20–30 times faster download rate compared to the original 3DGS method. The paper also includes a detailed evaluation of the proposed method, comparing it to previous solutions and showing its effectiveness in terms of memory reduction, rendering speed, and visual quality.The paper "Reducing the Memory Footprint of 3D Gaussian Splatting" by Panagiotis Papantonakis, Georgios Kopanas, Bernhard Kerbl, Alexandre Lanvin, and George Drettakis addresses the high memory requirements of 3D Gaussian Splatting (3DGS) for novel view synthesis (NVS). The authors identify three main areas contributing to the high memory usage: the number of 3D Gaussian primitives, the number of spherical harmonic (SH) coefficients, and the precision needed for storing Gaussian primitive attributes. They propose a comprehensive solution to reduce these memory requirements: 1. **Primitive Pruning**: An efficient, resolution-aware primitive pruning approach that reduces the primitive count by half. 2. **Adaptive SH Adjustment**: A method to choose the appropriate number of SH bands for each primitive, significantly reducing the overall memory footprint. 3. **Codebook-Based Quantization**: A codebook quantization method and the use of half-float representation to further reduce the stored size of the representation. These three components result in a 27× reduction in overall disk size and a 1.7× increase in rendering speed. The authors demonstrate their method on standard datasets and show that it significantly reduces download times on mobile devices, achieving a 20–30 times faster download rate compared to the original 3DGS method. The paper also includes a detailed evaluation of the proposed method, comparing it to previous solutions and showing its effectiveness in terms of memory reduction, rendering speed, and visual quality.