Better explanation of coordinates format in docs for faster+mask rcnn

[Robylyon93]

Change discussed in #1856 .

I propose to clarify the fact that ground-truth boxes in [x1, y1, x2, y2] format have:

• x-values between 0 and W
• y-values between 0 and H

The current description might be confusing since in general locations in a matrix are in format (row, col) instead of (col, row).

[fmassa]

Thanks!

Can you also send a PR for keypoint_rcnn.py?

[Robylyon93]

Sure! :)