Keras custom ImageDataGenerator to load images batch wise (with images segmented as patches)

Question

I have been trying to train a network consisting of a VGG16network followed by some LSTM layers. Since my images are quite big and because VGG16 do not scale with image size