DRosemei / RoMe

MIT License
220 stars 27 forks source link

Questions about "vertices_z" optimation #24

Closed luohongcheng closed 1 year ago

luohongcheng commented 1 year ago

Hello, in https://github.com/DRosemei/RoMe/blob/main/models/voxel.py#L313, the function ".detach()" is used, does it mean the value of vertices_z and the height MLP network would not be updated?

DRosemei commented 1 year ago

Hi, yes it is. Only update a small area according to the waypoint.

nkliuhui commented 10 months ago

请问怎么打开梯度传播,对MLP进行优化呢?对torch不太了解,改完无法确定对不对 with torch.no_grad(): [tab]self.vertices_z[activated_idx] = activated_vertices_z vertices_z = self.vertices_z.clone()