ScanNet classification, can't obtain 9,305/2,606 training/testing instances

[Yochengliu]

How did you fix the benchmark files to reach 9,305/2,606 instances, as you mentioned in issue #29 ?

Can you update the benchmark files with the fixed files (maybe scannet-labels.combined.tsv and classes_ObjClassification-ShapeNetCore55.txt in ./data_conversions)? Thanks.

[MingChaoSun]

Hi,

I think the cannet-labels.combined.tsv and classes_ObjClassification-ShapeNetCore55.txt in ./data_conversions are fixed files.

Thanks,
Mingchao

[Yochengliu]

Really? But I extracted the instances exactly following your instructions, and nearly got the same instances as #29 , i.e., 7814/2169, rather than 9,305/2,606.

I have checked the scannet-labels.combined.tsv and classes_ObjClassification-ShapeNetCore55.txt carefully, and I found the label name problem seemed to be still exist.

Is there anything I mistake? Thanks.

[MingChaoSun]

Sorry, I uploaded the wrong file before. T_T
I just updated the tsv file.

[Yochengliu]

Ok, I'll try it at once. Thank you very much!

[Yochengliu]

Using the updated tsv file, I get 9423/2634 instances rather than 9,305/2,606. It seems that there is something wrong, or are 9423/2634 instances also okay? Could you please have a check? Thanks.

[MingChaoSun]

This tsv file has been modified several times， so there is a small gap from the version one year ago. Compared with the original version, it should have corrected more label errors，therefore， a little more instances is reasonable.

[Yochengliu]

Ok, I get it. Thank you very much.