Exciting and interesting work! The mesh result of gaussian wrapping is impressive. But I have some questions about the rasterization. I notice that the median depth rendering is same as GGGS (both uses binary search except the transmittance computation), and the normal rendering is same as RaDe-GS and GGGS (both uses ray-space normal). So which is the key factor leading to gaussian wrapping outperforming the GGGS? Is the mesh extraction algorithm or the learnable normal property of gaussian model?
Exciting and interesting work! The mesh result of gaussian wrapping is impressive. But I have some questions about the rasterization. I notice that the median depth rendering is same as GGGS (both uses binary search except the transmittance computation), and the normal rendering is same as RaDe-GS and GGGS (both uses ray-space normal). So which is the key factor leading to gaussian wrapping outperforming the GGGS? Is the mesh extraction algorithm or the learnable normal property of gaussian model?