How to add conditional preprocessing with keras included functions?

Question

I have a dataset of images whith some of them that I need to flip before processing. I use the keras function flow_from_dataframe from the keras class ImageDataGenerator to get the train and validation DataFrameIterator but I can't find how to flip (not randomly) the images I want.

My dataframe has these three columns :

• filename : the filename of an image
• class : the target data
• should_be_flipped : a boolean that say if the image should be flipped before the training

Is there a way to achieve this woth keras included functions ? I saw that ImageDataGenerator has a preprocessing_function parameter but it doesn't allow to be conditional...