Open jiapei-nexera opened 2 years ago
I only implemented training PointNet with the full point clouds in the ModelNet dataset. If you want to train it with single-view or arbitrary-view point clouds, you can simply modify the code yourself or refer to their implementation (https://github.com/SMohammadi89/PointView-GCN).
But since PointNet is not goot at single-view / arbitrary-view point cloud classification, I would suggest you consider other sota models for a better performance.
Please do a simple comparison here https://github.com/SMohammadi89/PointView-GCN .
Is your implementation good at single-view classification? Or an arbitrary-view classification? Sometimes, when a single-view 3D point cloud classification is required, is there any parameter for us to specify?