I am not sure I understand how to tackle this question. What do they mean by delta?
Delta is commonly used to mean difference. So a “high delta of values” is a “high difference of values”
Ok, can you explain how can we think about this problem? Why is this wrong?
It might be helpful to work out an example and see what is going on. Usually you would convolve this filter with a larger matrix, but thinking just about the case where the matrix you are applying the filter to is also a 4x4 matrix can be helpful.
Think about a 4x4 matrix with a clearly defined horizontal edge and what value applying this filter would give you (I suggest actually writing this out). Did the different values in the filter really matter?
The idea here is that the values on the filter mean it will detect changes in certain directions better than others. The hint is suggesting that you think about connecting this with which direction you see the most change in values in and what happens if your other matrix has a change in that direction too.
Exactly. Prof Ng goes through the edge detection examples in some detail in the lectures. If you missed those points, it might be worth watching again and Martin’s great explanation will fit with what you see in the lecture.
Yea, he actually does a great job explaining that (as always). Especially towards the end when he explains the intuition of it. Not sure how I have missed that.
When I ran this exercise on a spreadsheet I can see it actually detect 2 different type of edges. Can I share this privately with you to understand what is it I’m doing wrong?