How can I reproduce the same image augmentation results when I\'m using the following data pipeline for an image segmentation task? I want to compare different models, so da
