search

CuiRuikai / Partial2Complete

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

Instructions for using the ScanNet dataset #36

Open PauliShit opened 2 weeks ago

PauliShit commented 2 weeks ago

Dear author, thanks for your amazing work！ How long will it take for you to release instruction for data preparation of ScanNet dataset. Your answer will be greatly appreciated

CuiRuikai commented 2 weeks ago

Sorry that I was busy with all kind of things after finished this project. But I may get time after the CVPR deadline and when I work on my final PhD thesis. If you are working on implementing this and encounter any problem. Please let me know and we could discuss to solve it.

PauliShit commented 2 weeks ago

Sorry that I was busy with all kind of things after finished this project. But I may get time after the CVPR deadline and when I work on my final PhD thesis. If you are working on implementing this and encounter any problem. Please let me know and we could discuss to solve it.

Dear author, the reason why I asked about the Scannet dataset is that I want to know whether your work could achieve multi-target point cloud completion in the same scene. I now have multiple depth cameras, but there is still a large part of missing after multi-camera point cloud fusion. I want to know whether your work could meet my needs. I am a novice in this area and don't know much . I would be most grateful if you could provide an answer.

CuiRuikai commented 2 weeks ago

Self supervised completion may heavily rely on priors about an object. Such a case is difficult for even supervised method. I think my method will also struggle. Also, my method seem not good at scene level completion as brought up by some issues.

If you have further questions to discuss, such as detailed research idea, it might be better to discuss via email instead of in public.