Terminology used is questionable, but conceptually it's not absurd.

If you train the model on a sample of 100 pink elephants bit only 20 blue elephants, it'll likely be better able to identify accurately pink elephants. Or if the nature of the lighting provides better contrast for pink elephants, etc.

I did some machine vision work back in the day, some objects are just easier to analyze than others for a variety of reasons, color being one.