comprehension issue

[TheJasal]

Hello. I an new to point cloud processing area. I want to know what is the difference between your method and data augmentation,please? Thanks.

[lijx10]

Data augmentation: randomly rotate the input point clouds. This method: for each input point cloud, rotate it acoording to the selected rotation group, do a max-pool on all outputs to get the final feature representation of that object.

[TheJasal]

Thanks for your reply very much!! When you test the experiment results, you said that during training phase you rotate the input point cloud in a 4 elements rotating group, by uniformly dividing 360 degree. Then if we manually rotate input point cloud by saying 60 degree during test phase, which is not in your rotation group, how does the network helps to improve the part segmentation actually?