Investigate and integrate the concept of robustness

[t-rad679]

This article discusses a notion called "robustness" of image features and that this is handled to differing degrees by different models. In particular, the article mentions that VGG based models are particularly good at extracting robust features, while non-VGG models are better at extracting non-robust features.

I don't fully understand this idea of robustness but would like to, and I would like to figure out how to integrate it into this project to allow for maximum diversity of output possibilities.

The article uses this library to explore robustness. It may prove useful.