Training of subcategories

[cjohn001]

Hello together,
I have a question in regards to training of an image classifier. I try do train a classifier which can deal with subcategories. I have images labeled with "apple", "apple granny" and "apple golden". I only want to have a single classification head (required to use classifier with mlkit) and I do not want to use multiple classifiers.
How would I have to train such a classifier and is there maybe an example available how to do it?
I am currently thinking about 2 possible approaches.

• Train with subcategories as additional main categories -> likely not a good idea
• Train by using each image with a subcategory twice, once with main group label "apple" and once again with label "apple-golden"

Would be great if someone could provide directions on how to best solve the task. Thanks for your help!

Best regards,
Christoph