Hello! Can the diffCD loss be directly applied to two corresponding clusters of point clouds?

Question

Hello! Can the diffCD loss be directly applied to two corresponding clusters of point clouds?

Opened this issue 3 days ago · 3 comments

Answer 1 · 2024-09-30T09:43:07.000Z

Hi, I'm not sure I understand the question. But if there are multiple clusters in the input point cloud the method should still work

Answer 2 · 2024-09-30T12:22:44.000Z

Thank you for your reply! I have a question regarding Equation (10) in the paper. It seems that this formula doesn't compute the distance directly between two point clouds like the Chamfer Distance (CD) does. Instead, it involves sampling from the implicit surface. Could you clarify if there's a way to adapt this formula to directly compare two point clouds without surface sampling? I'm looking for a loss function that works more similarly to CD, where the point clouds are directly compared.

Answer 3 · 2024-09-30T21:27:08.000Z

I see! DiffCD is inherently a method that minimizes the distance between a point cloud and an implicit surface, so it is not really suitable for matching two point clouds. Something like iterative closest point might make more sense in your case