Feeding Network with Random Image in Keras

I try to feed network with randomly chosen images. I have two kind of training data ( Base_Train(11k images), Augmented_Train(44k images) ) . Augmented_Train contains 3-5 images which have different perspective than Base images , for each image in the Base_Train. I want to feed network with Base_Train + randomly choosen one image from Augmented_Train for each image in the Base_Train for each epoch. I try to implement with loop. However, it was very ineffective. Randomly choosen image part took about 13 hours for each epoch. Therefore, this is not applicable.

I need an idea on how to implement this in efficient way. If you give me pieces of pseudocodes, I'm gratefull.

Thanks in advance.