Perpective warping in fastai

I have used data block api of fastai and applied augmentations like rotation, brightness, warping etc. Now I had a question that since we aren’t writing our own dataset class as in pytorch how do I avoid the black part in the image that comes after perspective warping and just take the pixels that aren’t zero? Can somebody please help me out?

Can someone please help out?!!

I still haven’t got any answer. Can you please suggest anything? @jeremy