I cannot reproduce the performance that you suggested on the ScanObjectNN with Point-PN.

[azhuantou]

Hi,

Thanks for your great work! I really like this idea and your brilliant paper-writing!

I reproduced the Point-PN on the ScanObjectNN by running the run_pn_scan.py with the default setting, but I got 85.3% accuracy (lower than the 87.1% claimed in the paper).

May I ask if I did something wrong?
Looking for your reply!

[liuhui0401]

Thanks for your interest. May I ask what type of GPU you used? Could you please provide your log file? Thanks~

[azhuantou]

Thanks for your quick reply.
The GPU is TITAN RTX.
The log file is here [out.txt]

[azhuantou]

Hello, @ZrrSkywalker @liuhui0401

Could you provide the log file, or open source the weight parameter file? I repeated the experiment many times, but still couldn't achieve the performance claimed in the paper.

Thanks!