Issue while Training

[AhmadAmmar]

trafficstars

Loss and accuracy parameters just say NaN

loss: nan - dice_coef: nan - dice_coef_rounded: nan - binary_crossentropy: nan - val_loss: nan - val_dice_coef: nan - val_dice_coef_rounded: nan - val_binary_crossentropy: nan

[Geoyi]

@AhmadAmmar, can you make sure that you're reading the data from the correct path?

[AhmadAmmar]

Yes, please see this "invalid value encountered in true_divide" error.

[Geoyi]

It may cause by an incorrect image you read in when it's at the batch normalization, see this stackoverflow issue.

I've seen this error before and vaguely remember that I did not read the dataset correctly, @AhmadAmmar. Please make sure either your RGB or label images aren't empty and can be viewed correctly.

And the label/mask data need to be in png format, see this line. If you want to use other image formats, please use this function in the utils.

Please also make sure your label data can be read in (256, 256) mentioned in another issue.