Course 3 Week 2 quiz; A bug in Autonomous driving quiz

Course Q&A Deep Learning Specialization Structuring Machine Learning Projects
1

I think, I found a small bug in my week2 quiz. I got a question (Q15) which is relatable question to the question above. I mean Q15 is based on a situation explained in Q14. However, the Q15 and Q14 which I received is no where related. Thankfully, my random choice of answer got correct and secured 100%, but this is the case to be taken care.
Please let me know, in case of any support.
Thank you :slight_smile: :slight_smile:

5

Same problem. Have a picture here.

shoot
shoot838×743 14.1 KB

6

Thanks for providing the concrete example of the problem Harsha mentioned. I agree that looks like a bug. I think the quiz framework allows random replacement of questions, so that you don’t necessarily see exactly the same questions each time you take the quiz. But if they are going to implement that kind of replacement scheme, then they’d better have a way to make it take into account the case that you have groups of related questions.

I will file a bug and let you know if the course staff have any feedback on this.

7

Thank you @paulinpaloalto :slight_smile:

8

image
image1621×634 37.9 KB

I have other question for quiz 2 is it for multi task we used softmax layer?

9

image
image1595×742 50.1 KB

is it end to end approach requires large amount of data?

10

Think about what softmax does: it chooses which one of the K possible classes is most likely what is in the image. So it is used in a multiclass classificaiton, e.g. there is one of ten possible types of animals in an image, but only one.

In this task, there could be any number of different types of signs and traffic lights in the image and we want to enumerate all of them. Think about how you would represent that output.

2 Likes
11

Yes, the end to end approach would require a lot of data. But they are not telling you how much data you have here. The question is when the two step cascaded approach would be better. Well, if you had a pretrained model that could localize objects in an image (maybe including lots more types of objects than signs e.g. like YOLO), then you could use that as the first step and build a binary sign classifier that can distinguish stop signs from other signs based on having the isolated images of the signs. Or maybe you get lucky and you can find a pretrained model that does the second step as well and you just have to create the framework to construct your solution based on the other two models.

2 Likes
12

Ok thanks sir!