KerasDataGenerator for DataPipeline

[Salehoof]

Hi, I am working on an image change detection problem using keras. The problem dataset consists of 3 images for each scene: The reference image, the moving image (which should be co-processed with the reference image to find changes) and the ground truth image. The augmentation should be done the same for these three images for each scene. I want to use Augmentor for this dataset and I have found there is a DataPipeline class that can be used as a keras data generator, But I want a class that can accept the path of the images grouped in a list as input not the images as an array-like matrix. (This is needed for huge-size datasets that can't directly fit into memory.) Is there a way to do so? Thanks in advance.