jiawei-ren
commented
7 months ago Hi, we followed MAV3D (see Section 4.3 Real-time rendering). Although it could be harder to optimize, sharing a texture map across a deformed mesh sequence would be indeed more ideal and interesting to try.
Based on your paper, mesh and texture map are extracted and optimized across each timestamp individually. However, I'm curious whether it is possible to extract mesh and texture map once with static DreamGS, and realize motion of characters by optimizing only vertices of mesh for each timestamp with the texture map intact. I think it is more natural because texture map of a character(object) should be unchanged for rigid motions.