I have a feeling that something is messed up in the 9th question of week’s 2 quiz (Depthwise Separable Convolutions). Whatever I select, the grader always shows 0/1 (not even 0.1/1). Some feedback would be helpful.
I have the same problem.
You do realize that it’s a “select all that apply” question, right? And there are 8 possible “yes/no” items. That’s a total of 256 possible combinations and I don’t think the grader gives partial credit.
Don’t get me wrong: I feel your pain. I have not managed to get it right yet either … But I’m not ready to blame the grader yet
@ paulinpaloalto The grader does give a partial credit for all other questions.
Finally I found all correct answers.
Some choices about shape is confusing for me, probably because of my poor English and lack of understanding of the notations…
I have also managed to get the correct combination of answers, but can’t agree on one of them. Namely, the grader expects us to select “the final output has a shape Nout x Nout x C’, where C’ is the number of channels in the previous convolutional step”. This is wrong, because the final output of the depthwise separable convolution has the number of channels equal to the number of pixelwise filters. It can also be seen in the next question, by the way.
Agree with @pugach on the option selection pertinent to final output. Please update us on this question before I return to the quiz.
@Nicholaskaixiang I have the same problem, I think it’s a typo because the output filter for a depthwise convolution is nout x nout x nc, not n’c (which is the final result)
Edit: I think I know the answer. This question is not asking about the depthwise step, but the whole Depthwise Separable Convolution, which also includes the point wise step. That’s why n’c is the correct size for the output image.
I think it should have an advice in this question saying it is different from the others that gives partial credit. It would give the opportunity to be extra careful and it wouldn’t give a room for doubts when you not receive a partial credit for the items you got right.