Thing is when fastai crops images it some times takes away some landmarks (there are total of 9 landmarks in every image) by croppping it. So This leads me into a problem when making a databunch because the target sizes are different . So Is there any way I can fix this ?
No it’s not, I have my own study group I run (Walk with fastai2) where I show examples not covered in the course. The closest to it is the Headpose example