返回总结列表https://blog.csdn.net/albert199199/article/details/103531486
KPConv: Flexible and Deformable Convolution for Point Clouds
Key points:
-
3D conv (kernel): location correlation*Weights
-
L_fit: kernel location loss
-
L_rep: for repulsiving
-
3D shape classification: ModelNet40
-
Part segmentation: ShapenetPart
-
3D Scene Segmentation: Scannet (indoor), S3DIS (indoor), Semantic3D (outdoor) and Paris-Lille-3D (outdoor)
-
code