ma-xu / pointMLP-pytorch

[ICLR 2022 poster] Official PyTorch implementation of "Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

GEOMETRIC AFFINE MODULE

AArutoria opened this issue · comments

May I ask what part of the GEOMETRIC AFFINE MODULE code is? Looking forward to the author's answer

commented

@AArutoria Hi, I remember that I responsed to you. The code is here:

if self.normalize is not None:

Please let me know if any further questions.

@AArutoria Hi, I remember that I responsed to you. The code is here:

if self.normalize is not None:

Please let me know if any further questions.

Thanks to the author for answering questions. I have a few more questions. The first point: Figure 2 shows a stage of PointMLP, which shows that the point cloud is directly input into the GEOMETRIC AFFINE MODULE. But in Figure 6, the point cloud enters stage1 after the Embedding operation. Does this mean that both point clouds and embedded point clouds can be input into GEOMETRIC AFFINE MODULE? Second point: In Figure 6, I can't see where the GEOMETRIC AFFINE MODULE is. Is it in Embedding or in stage1? Thanks again for the author's answer

commented

@AArutoria

  1. yes. BTW, the inputs in figure 2 actually are features. we plot as points for better visualization. You can refer the codes for better understnading.
  2. It is in the aggregation. Please correct me if I remember wrong.

Feel free to ask me if you have any further questions.

@AArutoria

  1. yes. BTW, the inputs in figure 2 actually are features. we plot as points for better visualization. You can refer the codes for better understnading.
  2. It is in the aggregation. Please correct me if I remember wrong.

Feel free to ask me if you have any further questions.

Thanks for your patient answer. I have a new understanding of your code