2 error messages are not very explicits


#1

For a big dataset when I set “Load dataset to memory” to “Full Dataset”, there’s an error when the training is started.
29
That message disappear when “Load dataset to memory” is set to “One batch at a time” but then another message is displayed:
31
This is due to the fact that the image in the dataset don’t have the same dimensions and do not have the same Height/Width ratio. After using rescaling to 224 224, it works OK.

Both message are not very explicit, would need to be stated differently.


#2

Yes. We definitely have much to improve on error reporting. Currently we pass the error message from the DL framework as it is. This may not be very familiar to the users who are new to Deep Learning.

The reason behind two error messages is that there are 2 processes (one generating batches and other reading batches). First error message is from the reader process (because it didn’t find next batch) and 2nd one is from the generator process which had issue in generating the batch.


#3

OK, thanks for the detailed explanation. Always best to understand what is going on underneath.