Trying to understand non-max suppression (week 3 quiz question 6)

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

Hello,

I’m trying to figure out why I don’t seem to get question 6 of the third week quiz right. Is there some sort of bug, or do I simply not understand non-max suppression? The question is

imagem
imagem771×564 113 KB

I would say the car with bounding box with pc = 0.26 is discarded, and then since the two remaining boxes have an IOU above 0.5, only the one with pc = 0.73 “survives”. Therefore the answer must be “True”.

However, I am given 4 choices, and none of the two options containing “True” have been accepted as correct after I repeated the quiz. Does this mean that two boxes remain after non-max suppression?

I’m not sure posting the exact possible choices goes against guidelines but I can also provide them.

Thank you,
Afonso

2

I’m looking at the copy of the quiz that I passed, and there it only has the text above the image, where it asks how many boxes remain. It doesn’t have the lines of text under the figure that discuss only the boxes for cars as a True/False question.

I’ll forward this to the course staff.

1 Like
3

Hi @Afonso_Rodrigues,

One of the incorrect answers was accidentally marked as the “correct” one. I have now fixed the mistake.

Thanks you for bringing this to our attention.
Mubsi

1 Like
4

Hello @Mubsi.
Thank you for your consideration.

But I think there is still the same problem.

According to my understanding, pc = 0.26 is discarded due probability 0.4 threshold, and among the remaining two cars, the car with a maximum probability of 0.73 is selected, and the car with the probability of 0.62 will be discarded because IoU is greater than 0.5.

But in the question, there are two ‘True’ answers but both reasons do not match with the above explanation.

Did the error not be corrected? Or is my understanding wrong?

Thank you.

5

This looks like what @paulinpaloalto handled recently. Here is the link.

6

Yes, I agree with Nobu’s analysis: it’s exactly the same problem that was reported on that other thread. Your reasoning agrees completely with my understanding of the correct answer there. The bug is filed as of 3 days ago, but I haven’t gotten any response on that yet. Sorry, but you’ll just have to experiment to find out which of the wrong answers the grader actually accepts for now. :nerd_face: