question about training data for faster-rcnn

[jun wei Chen]

i have a weird question want to ask , i have trained a network using py-faster-rcnn , 

i download the  dataset  from internet , which is including annotation file 

so the network know  the region proposal is foreground or background  ,

but if my dataset are all composed of images  which is croped image , no background at all ,

then i set the axis to :

xmin = 0

ymin = 0

xmax = image.width

ymax = image.height

can the network learn what is the background  ?