What is the best way to create augmentation on image dataset while training instance segmentation?

Question

Background : I am using YOLACT instance segmentation model to train set of images. The dataset size is very small (~20 images). The model doesn\'t converge properly (of-cour