What makes an augmentation a good augmentation?

Ongoing Project

Abstract

In this work, we analyze image augmentations under various image quality assessment constraints and try to model a space of all high-performing augmentations used. We further aim to deduce that for an augmentation to perform better than baseline, it needs to fall within this closed bound space, however, we clarify that not all augmentations within this closed bound space qualify to perform better than baseline.

Participants