Question about the dataset split for training and testing

[aflyingnoob]

Hello, thank you very much for sharing your code and dataset! While reviewing the documentation and code, I have a question that I hope you can clarify:

How are the training and testing sets of the dataset split? Specifically:

1. Are they split based on seen and unseen categories?
2. Or are the training and testing sets completely independent with no overlap?
   If possible, could you please provide more details about the splitting criteria or method? I would greatly appreciate it!

Looking forward to your reply. Thank you! 😊