search

CVMI-Lab / PAConv

(CVPR 2021) PAConv: Position Adaptive Convolution with Dynamic Kernel Assembling on Point Clouds
Apache License 2.0
287 stars 40 forks source link

Why set npoints=None for Sampling layer in pointnet2_paconv_seg.py? #41

Closed whuhxb closed 2 years ago

whuhxb commented 2 years ago

Hi @mutianxu

I go through the seg code, and have one question. Why the points for sampling layer is set to None for paconv? Do you mean setting the npoint to the original point number in all the sampling layer?

In the paper, it's said that replacing MLP with PAConv. But, actually, PAConv is used in sampling layer?

https://github.com/CVMI-Lab/PAConv/blob/main/scene_seg/model/pointnet2/pointnet2_paconv_seg.py#L28

mutianxu commented 2 years ago

"None" means not specify the concrete number of points after downsampling. When it sets to "none", the points are downsampled by 1/4, as here

I would be appreciated if you can read our code logic very carefully before opening an issue.

Thanks a lot! Mino

whuhxb commented 2 years ago

Hi @mutianxu

Thanks a lot for your kind answering.