Projecting the output of NonRigidPerPixelSE3

[lolz0r]

Hello! I am using test_warpingSE3.lua as a basis for experimentation. If I provide NonRigidPerPixelSE3 a BxWxHx3x3 tensor (3-rot, 3-trans, per pixel) it outputs a tensor in the form BxWxHx3x4 (which is likely a formal 3x4 transformation matrix)

The problem I have is I am not sure if gvnn provides a transform layer for data in this form. Transform3DPoints_depth, Transform3DPoints_r, Transform3DPoints_rt do not seem to be up to the task?

How does one utilize the output of NonRigidPerPixelSE3?

Best wishes, Michael