mmgeneration icon indicating copy to clipboard operation
mmgeneration copied to clipboard

Published 20 hours ago • verified publisher icon open-mmlab

Why does train.py build a val_dataset using train pipeline?

Open gvalvano opened this issue 3 years ago • 5 comments

Why do you use the train pipeline for the validation set at this line of code?

To my understanding, the line of code:

val_dataset.pipeline = cfg.data.train.pipeline

should be

val_dataset.pipeline = cfg.data.val.pipeline

Is there something I don't understand?

gvalvano avatar Feb 03 '22 17:02 gvalvano

Hey @gvalvano, thanks for your interest in our project. This seems a mistake, and we will fix this soon. We would also be glad that if you could create a pull request to fix this.

LeoXing1996 avatar Feb 04 '22 02:02 LeoXing1996

Hi @LeoXing1996 Thanks for the help. Sure, I'll be happy to do so!

gvalvano avatar Feb 04 '22 13:02 gvalvano

Opened pull request #244

gvalvano avatar Feb 05 '22 13:02 gvalvano

Hi @LeoXing1996 it seems that the pull request failed during the "lint" process. I don't have familiarity with this. Should I take any action to address this issue?

gvalvano avatar Feb 12 '22 11:02 gvalvano

Please feel free. That lint error is caused by the network error. I just rerun the CI and all checks are passed. We will merge this in this version.

LeoXing1996 avatar Feb 12 '22 12:02 LeoXing1996