Generation of testing data on ModelNet

[qinzheng93]

Hi Liang,

I notice the testing data on ModelNet is already preprocessed, so I am wondering how do you generate them. Is it generated using the same code as the training data with the seed fixed? That is to say, the same 1024 points are sampled first, and cropped to different point clouds and transformed separately.

As far as I know, sampling different 1024 points for the source and target point clouds can significantly affect the performance, which is the original implementation in RPM-Net. How does GMCNet perform in this setting?

Thanks,
Zheng