Question about Data augmentation?

I’m doing data augmentation to reduce overfitting

I’ve used different techniques and due to the fact i’m using my classifier after a pipeline of object detector and tracker, so usually there’s a random crop of the actual object, that’s why i added data augmentation with random zoom

I’m asking about the black border, would it affect the training badly? or as it’s done in both classes, it would be neglected by the network?