The hyperparameters using in the simulation experiments

[HXYfighter]

Excuse me, I've been trying to reproduce all of the experiment results in your PointNetGPD paper recently. When running the main_1v_mc.py, it seems that keeping all the hyperparameters the same as those default values except batch-size(which I set to 32, 256 and 1024), all the test accuracy is lower than the result in your paper(79.45%) by more than 5 percent. I'm wondering what's wrong with my operation, maybe some hyperparameters need to be changed in this experiment? I'm sincerely looking forward to your guidance.

By the way, I'm also from a group in Tsinghua University(in EE Department). We have also bought all the facilities for this experiment, so I'll be especially grateful to you if we can keep in touch. My e-mail: [email protected]