Open wangyuanbiubiubiu opened 7 months ago
In this part of the code, sdf loss uses the sdf gt supervised by the net-unit space, but the part used for eikonal supervision specifically transfers the gradient of sdf from the net-unit space to the obj-unit space. What is the reason for this? Can not the net-unit space meet the eikonal constraint? https://github.com/PJLab-ADG/nr3d_lib/blob/5398efb16b87116cc3c9e3ad588c537d1916e877/models/fields/sdf/utils.py#L253
In this part of the code, sdf loss uses the sdf gt supervised by the net-unit space, but the part used for eikonal supervision specifically transfers the gradient of sdf from the net-unit space to the obj-unit space. What is the reason for this? Can not the net-unit space meet the eikonal constraint? https://github.com/PJLab-ADG/nr3d_lib/blob/5398efb16b87116cc3c9e3ad588c537d1916e877/models/fields/sdf/utils.py#L253