I’m having one heck of a time trying to understand how does a final output of a convolution look like and work.
In lesson 6 Jeremy is talking about a matrix that is 11x11x512. That matrix is an output of a convolutional part of neural network that we worked on during class. We average across that matrix and get a heatmap out of it.
So from my understanding 512 number comes from number of filters (aka kernels) from the previous layer, is that right? Jeremy says each of those 512 “layers” are features.
Part that I don’t get is what do those features represent? For example if we have a dog classifier would the first layer represent how “hairy” the image that’s the input is? Am I thinking about this right?