CuiRuikai / Partial2Complete

[ICCV 2023] P2C: Self-Supervised Point Cloud Completion from Single Partial Clouds
MIT License
157 stars 9 forks source link

question about my own dataset #22

Open YunLixiang opened 5 months ago

YunLixiang commented 5 months ago

Hello author, I have great admiration for and interest in your achievements. Now I am ready to use your advanced framework to try to repair some of my own point cloud data. May I ask whether your framework supports point cloud data repair with a large amount of data? My data is about 100,000 points, and my point cloud data is about ground point cloud data of mining areas. Can this type of data amount be used for testing? Thank you very much for your answer!

CuiRuikai commented 5 months ago

No, we only evaluated on point clouds that are much sparser than yours. That is 2048 points. If you need to generate dense point cloud, you have to design your own decoder as ours is just a simple MLP. Scaling the MLP to millions of points will take too much memory and computation.

Sent from Outlook for iOShttps://aka.ms/o0ukef


From: YunLixiang @.> Sent: Monday, June 10, 2024 3:23:18 PM To: CuiRuikai/Partial2Complete @.> Cc: Subscribed @.***> Subject: Re: [CuiRuikai/Partial2Complete] question about my dataset (Issue #22)

I am going to use your advanced framework to try to repair some of my own point cloud data. May I ask whether your framework supports point cloud data repair with a large amount of data? My data is about 100,000 points. May I ask whether this level of data can be used for testing?

— Reply to this email directly, view it on GitHubhttps://github.com/CuiRuikai/Partial2Complete/issues/22#issuecomment-2157293301, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ALJ7PYUHZWGDELMFB4QL6YLZGUZ4NAVCNFSM6AAAAABJBSEF32VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNJXGI4TGMZQGE. You are receiving this because you are subscribed to this thread.Message ID: @.***>

YunLixiang commented 5 months ago

Perhaps I should first make an affine transformation of point cloud data to narrow down my scene, and then conduct relevant tests. If there is a better memory computing network in the future, my data can be applied to actual production. Thank you very much for your reply, which is of great significance to a beginner of point cloud deep learning

CuiRuikai commented 5 months ago

It would be feasible if you first downsample the point cloud and generate only a coarse shape, then use some upsampling methods to recover the full point cloud.

Sent from Outlook for iOShttps://aka.ms/o0ukef


From: YunLixiang @.> Sent: Monday, June 10, 2024 4:41:17 PM To: CuiRuikai/Partial2Complete @.> Cc: CuiRuikai @.>; Comment @.> Subject: Re: [CuiRuikai/Partial2Complete] question about my own dataset (Issue #22)

Perhaps I should first make an affine transformation of point cloud data to narrow down my scene, and then conduct relevant tests. If there is a better memory computing network in the future, my data can be applied to actual production. Thank you very much for your reply, which is of great significance to a beginner of point cloud deep learning

— Reply to this email directly, view it on GitHubhttps://github.com/CuiRuikai/Partial2Complete/issues/22#issuecomment-2157449796, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ALJ7PYTBMGQW336PCE7FS43ZGVDA3AVCNFSM6AAAAABJBSEF32VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNJXGQ2DSNZZGY. You are receiving this because you commented.Message ID: @.***>