IoU of pefectly overlapped bounding boxes is not 1

[rttus]

Hello,

I am trying to understand how bev_box_overlap works. Something i noticed is when i pass in the same bounding boxes for detected and ground truth, i expect a IoU of 1 but it turns out to be 0.33. Wondering if my understanding is right!

`
dt_boxes = np.array([[-4.31, 23.72, 1.6 , 0.62, -5.28]])

gt_boxes = np.array([[-4.31, 23.72, 1.6 , 0.62, -5.28]])

overlap = bev_box_overlap(dt_boxes, gt_boxes)

print(overlap)
`

In this case i expect a IoU 1 because both boxes are essentially the same and they overlap like 100% instead i get a IoU of 0.33333257.

I tied to understand how quadrilateral_intersection is implemented and looks like one of the corner in the box1 is falling outside the corner of box2
###########
-5.001483
24.227888
Inside
###########
###########
-4.478707
24.561211
Inside
###########
###########
-3.618516
23.212111
Ouside
###########
###########
-4.141293
22.878788
Inside
###########

This gives area of intersection of polygons to be 0.495999 where as the expected is 0.992000!
Can someone please clarify?