dtld123
dtld123
https://x.yynet-1.site/auth/register?code=y2Kn 用了三年了，开了就不用管
Yes, I can only get about 10% accuracy on the ScanObjectNN dataset. I don't know where the error is.
Thanks for your reminder. I tested the results using the cat_name_pt_feat.npy and xyz_label.npy files you provided. The OpenShape-SparseConv + ScanObjectNN result is 48.93, which is slightly lower than the 56.7...
I understand that the simplest slice is main_split_nobg/objectdataset. Could you please help check it?
This is very strange. The number of test sets given by OpenShape is 2890, but none of the partitions of scanobjectnn have this number.