Not enough images in validation set in AutoML when done automatically

Question

I'm trying AutoML from GCP and I'm stuck at training: I get an error message saying there is not enough images having the "Italy" tag in the validation set. I have uploaded the images manually, and so I did not specify which image goes to which set (training, validation, test).

The weird thing is, "Italy" has more images than the other tags, so why is there a problem with it ? Can I fix it from the AutoML interface or do I really have to upload a CSV to fix the bug ?

I already tried to add some new images with the "Italy" tag and it didn't change, I still get "0" in validation...

If it helps, an image can have multiple tags (in my example: one sport + one country).

Answer 1

You can resolve that issue by manually reassigning some images to the VALIDATE dataset, see this answer.