search

HpWang-whu / YOHO

[ACM MM 2022] You Only Hypothesize Once: Point Cloud Registration with Rotation-equivariant Descriptors
https://hpwang-whu.github.io/YOHO
134 stars 19 forks source link

Generalization and scale differences #10

Closed fdila closed 1 year ago

fdila commented 1 year ago

I am reading about the generalization capabilities of the backbone.

"Assume the scale difference s between DS and 3DMatch. FCGF is a spatial-voxelization encoder and we set voxel size to 0.025m for 3DMatch in training, thus the voxelized DS should have an similar scale by voxel size a~0.025*s"

What do you mean exactly by the scale difference s?

HpWang-whu commented 1 year ago

Hi @fdila, It means the approximate ratio between the range of your scans over the average range of the 3DMatch scans (~4 meters).

Yours,