Describe how convolution works. What about if your inputs are grayscale vs RGB imagery? What determines the shape of the next layer?