@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.