Issues of Equivariant KPconv
Closed this issue · 1 comments
JudgeLJX commented
Hello,
Thanks for your contribution.
I have an issue:
How did you convert the kpconv to an equivariant version, is there any changes you made to their structure besides the quaternion representation?
Best Regards,
Judge
nintendops commented
Hi Judge,
we convert kpconv to the rotationally equivariant version by rotating the "kernel points" X times and extracting these rotated features into an extra dimensions. Please refer to the paper for the details.