C1_W3_Lab_1 convolution visualization

Hello everyone, I’m at week 3 lab visualization portion but I’m not sure what similarity we’re supposed to see between the visualizations of 0, 23 and 28. There is some similarity if we only look at the first 2 columns but columns 3 and 4 I am not sure what common feature we’re supposed to see. Any help?

I believe that as the layers get deeper, the NN starts to decipher more and more complicated features which are formed from the basic ones.
So I think there would be a similarity between 23 and 28; I am not sure about 0.
@balaji.ambresh could you confirm if I am right and help us further cause I am also not able to understand why 0 is to be compared with 23 and 28
Thanking you
Mayank Ghogale

@MayankGhogale Your explanation of conv is correct. To iterate, as you go deeper in the NN, conv filters are capable of figuring out more complex shapes.

The visualization is shown for 3 different images. There is no relationship between images. The notebook author shows 3 images as an example to show that the conv behavior that is observed isn’t pure coincedence.

Please see this post.

Thanks @balaji.ambresh
I hope that helps @annena