xingyul/flownet3d

Details on the Scan Registration Application

ymxlzgy opened this issue · 0 comments

In the Supplementary of your paper, E part describes the details on the Scan Registration Application saying that " we iteratively regress twice for the scene flow (i.e. predict a flow from point cloud 1 to point cloud 2, and then predict a second residual flow from point cloud 1 + first flow to point cloud 2) " . I want to know what is the "residual flow" . Did you just estimate scene flow twice? And I want to know the detail of how to translate the scene flow into the rigid motion estimation, how you " fit a rigid transformation from the point cloud 1 to the point cloud 2 + scene flow, as they have one-to-one correspondences " ? Is the code of this part available ?