The paper "Blur Agnostic Gaussian Splatting through Multi-Scale Kernel Modeling" addresses the robustness of Gaussian-Splatting-based methods against various image blurs, such as motion blur, defocus blur, and downscaling blur. These methods often overfit and produce poor results under these conditions compared to Neural-Radiance-Field-based methods. To tackle this issue, the authors propose Blur Agnostic Gaussian Splatting (BAGS), which introduces a Blur Proposal Network (BPN) to model per-pixel convolution kernels and a quality-assessing mask for regions of blur. BPN considers spatial, color, and depth variations of the scene to maximize modeling capacity. Additionally, BAGS employs a coarse-to-fine kernel optimization scheme, which gradually increases the resolution of the training images and estimated kernels, improving stability and efficiency. The method is evaluated on three image blur scenarios and shows significant improvements over existing approaches, achieving photorealistic renderings under challenging conditions. The contributions of the paper include the introduction of BPN, a coarse-to-fine optimization scheme, and the demonstration of BAGS's effectiveness in various degradation scenarios.The paper "Blur Agnostic Gaussian Splatting through Multi-Scale Kernel Modeling" addresses the robustness of Gaussian-Splatting-based methods against various image blurs, such as motion blur, defocus blur, and downscaling blur. These methods often overfit and produce poor results under these conditions compared to Neural-Radiance-Field-based methods. To tackle this issue, the authors propose Blur Agnostic Gaussian Splatting (BAGS), which introduces a Blur Proposal Network (BPN) to model per-pixel convolution kernels and a quality-assessing mask for regions of blur. BPN considers spatial, color, and depth variations of the scene to maximize modeling capacity. Additionally, BAGS employs a coarse-to-fine kernel optimization scheme, which gradually increases the resolution of the training images and estimated kernels, improving stability and efficiency. The method is evaluated on three image blur scenarios and shows significant improvements over existing approaches, achieving photorealistic renderings under challenging conditions. The contributions of the paper include the introduction of BPN, a coarse-to-fine optimization scheme, and the demonstration of BAGS's effectiveness in various degradation scenarios.