land_cover_classification_unet icon indicating copy to clipboard operation
land_cover_classification_unet copied to clipboard

Published 20 hours ago • verified publisher icon srimannarayanabaratam

Bug in Loss/train writer

Open SergioPovoli opened this issue 2 years ago • 0 comments

Hello sir, I found your work thanks to the awesome Medium post and first of all I congratulate you.

During my code exploration I think I found a bug in line 138 of train.py :

writer.add_scalar('Loss/train', epoch_loss/n_train, epoch+1)

I think in fact that the division is wrong because it must be divided by the number of batches and not by the number of samples of trianing.

Am I wrong? Or is it really an error ?

SergioPovoli avatar Mar 21 '22 14:03 SergioPovoli