Misleading coordinate for bounding boxes

I am trying to use the brambox.io.parser.annotation.DarknetParser parser from brambox on an already existing dataset composed of images (.png) and annotations (.txt).

The boxes drawn at the moment are not correct and I am digging to understand why. In the meantime, I stumbled across this line in brambox/util/_visual.py :

pt3 = (int(box.x_top_left + box.width), int(box.y_top_left + box.height))

I understand that therefore box.y_top_leftrather represents box.y_bottom_left since one is adding the box height or am I misunderstanding something ?