the size of the augmentation will adjust to the physical size of the image target automatically. Specifically, the view matrix we provide creates an Euclidean coordinate system where 1 virtual unit is equal to the longer side of your image target.
So, if you had a cube with a side length of 1 virtual unit as an augmentation, it would have a 10cm side length on a 10cmx5cm target, and a 20cm side length on a 20cmx20cm target.
The size in pixels is entirely irrelevant for this process. The physical size of the image (how big you print it, how big it is shown on a screen) is what matters.
Does that clear things up?