ZrrSkywalker / Point-NN

[CVPR 2023] Parameter is Not All You Need: Starting from Non-Parametric Networks for 3D Point Cloud Analysis
MIT License
455 stars 49 forks source link

Using PointNN for Detection #19

Open mysticalfirellama opened 1 year ago

mysticalfirellama commented 1 year ago

Hi,

The paper / main diagram depicts four use cases: Classification, Few-shot Learning, Segmentation, Detection. These make PointNN appealing due to its versatility, especially without needing parameters.

I can find examples for classification and segmentation in the code, but not for detection. Will a detection example be made available?

Thanks,

mysticalfirellama commented 1 year ago

Update: I've just read the detection section of the paper, and it states that "we utilize two popular [parametric] 3D detectors" for region proposals.

It makes sense why these (parametric) detectors are not bundled with PointNN. Although, the diagram is slightly misleading as it appears to show that parameter-free detection is supported.

bobo-liang commented 10 months ago

同求,检测,会更新吗?