Transpose convolution confusion

Hi there,
Having followed the lecture, I can’t understand how the output in the quiz is as it is.
(padding = 1, stride = 2)

It seems to me as if in order to do the exercise we could make the analogy from the lecture slide which has the same dimensions of input, filter, output, padding and stride.

So by analogy if this is the input

we first multiply this “1” by each of the following numbers:

because placing the filter on the output, we ignore the paddy area. I would think (but of course I am mistaken) we would get

which is obviously a mistake.
Having watched the lecture again and again I can’t see the problem.

The point is that’s not the only input at that point in the output. With stride = 2, that position also overlaps with values from the first horizontal stride and both of the first two vertical strides, right? So you need to watch more than just the first step in the video.

In other words, that position in the output is the sum of 4 separate values from 4 separate steps of the transpose convolution, not just the first step, which is what you showed. This was clearly spelled out in the lecture video if you watched it all the way through.

1 Like

Oh, right! The overlapping later adds to the value of that cell. (Don’t worry I did watch everything a few times, I was just so disapointed after the first step, that my brain stopped :smiley: Thank you.)
Though I passed the quiz, I now did this exercise once again with a quiet mind, just to make sure, and it was fine.