Quiz question 6

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hello,

Can someone explain me the right answer for the 6th question of this quiz ?
(The one with the number of boxes remaining after non-max suppression)

First I discard all boxes with probability < 0.4, I remove 1 box (the 0.26 car), then
I take the box with max probability (0.94 pedestrian), check IOU, and keep this box (no intersection nor union) ==> 1 box
I take the next box with max probability (0.74 tree), check IOU, seems < 0.5, so remove that box.
Next : 0.73 car, IOU > 0.5 with 0.62 car, keep 0.73 car and remove 0.62 car ==> 2 boxes
Next : 0.58 motorcycle, no intersection nor union, keep that box ==> 3 boxes
Next 0.48 tree, no more intersection nor union (0.74 tree has been removed), keep that box ==> 4 boxes.

This is not the right answer but I don’t understand why.

Thanks

2

The IoU threshold means that when 2 boxes have value of IoU > 0.5, then you need to remove them.
When you have IoU < 0.5, it means that the 2 boxes have lesser intersection area, hence we assume them as 2 different boxes. So, it will be 5 boxes in all

3

Hello, thanks for this answer.

When you say “you need to remove them”, you mean you only remove the lower “Pc value” box ?

The remainig boxes are :
motorcycle (no intersection)
pedestrian (no intersection)
0.73 car (0.62 car removed because IoU with 0.73 car > 0.5)
0.74 tree
0.46 tree ?

4

Yes you are absolutely correct that we need to remove the box corresponding to the lower probability, and indeed these will be the 5 remaining boxes