ma-xu/pointMLP-pytorch

LocalGrouper is different from the description in the paper

Closed this issue · 4 comments

Qianyu1998 commented

Thanks to the author for the code. I would like to ask a question, the variable σ is calculated in the code is not the same as the description in the paper.

ma-xu commented

Thanks to the author for the code. I would like to ask a question, the variable σ is calculated in the code is not the same as the description in the paper.
@Qianyu1998 Thanks for your interest. Could you please point out where the difference is? in case there are some misunderstandings or I can fix it.

Qianyu1998 commented
ma-xu commented

@Qianyu1998
For \sigma, please double-check, looks like there is no differences.

I assume you are talking about the refine module. The motivation is that we should provide a simple method that can avoid complex local computations, operations like normalization perfectly match this requirements.

martin-liao commented

Thanks for your interpretation! I guess the affine module plays the role of spatial transformer in PointNet ,which maps input features to a canonical space,right?(Yeah it seems that more relative to normalization...)

@Qianyu1998 For \sigma, please double-check, looks like there is no differences.

I assume you are talking about the refine module. The motivation is that we should provide a simple method that can avoid complex local computations, operations like normalization perfectly match this requirements.