search

erikwijmans / Pointnet2_PyTorch

PyTorch implementation of Pointnet2/Pointnet++
The Unlicense
1.45k stars 332 forks source link

point clouds with various size #50

Open dandan19793 opened 5 years ago

dandan19793 commented 5 years ago

Hi Erik,

is it possible to use point clouds with various size? I am struggling to understand if the sampling and grouping step allows me to do that.

Thanks

erikwijmans commented 5 years ago

What do you mean by various sizes? If you mean various numbers of points, then yes (although you'll need to pad/duplicate points for batching).

If you mean various spatial sizes, also yes, but you mean need to change the sizes of the ball queries in the grouping stages to get good performance.

dandan19793 commented 5 years ago

thanks for replying so quickly. I have different numbers of points in each point cloud ranging between 500 and 120k. Do you mean i should decide on a fixed size input (lets say 10k) and then pad/duplicate or downsample my whole data set to 10k?

On Mon, 27 May 2019 at 21:39, Erik Wijmans notifications@github.com wrote:

What do you mean by various sizes? If you mean various numbers of points, then yes (although you'll need to pad/duplicate points for batching).

If you mean various spatial sizes, also yes, but you mean need to change the sizes of the ball queries in the grouping stages to get good performance.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/erikwijmans/Pointnet2_PyTorch/issues/50?email_source=notifications&email_token=AENNRZMOLMGFWBZ2YJBF3XTPXQ2HXA5CNFSM4HP5XOQKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODWKNF7I#issuecomment-496292605, or mute the thread https://github.com/notifications/unsubscribe-auth/AENNRZKYSIQCIAIKFJZM5GLPXQ2HXANCNFSM4HP5XOQA .