How can i do many things to configure data with keras

Question

I am a beginner learning deep learning by Keras. The ImageDataGenerator class in Keras and the flow_from_directory function made it easy to label images. But all of themAs I configure the electory, there seems to be a limitation in configuring the dataset.

For example, if model want to learn ensemble by dividing the data into 5 folds, then should i make many directory?

or to bootstrap the data and use the verification data as out of bag data, should I implement it myself?

I'd appreciate it if you could give me a little advice!

As a reference photo, I uploaded the process of labeling with the flow_from_directory function, (88 state classification model)