Problem with deep feedforward networks

In Chapter 2, Deep Feedforward Networks, we learned to identify (classify) images of fashion items using deep feedforward networks. The size of each image was 28 x 28 and we connected one neuron to each pixel. This way, we have 28 x 28 = 784 neurons in the first layer itself. But in the real world, images are barely this small. Let's consider a medium-sized image of size 500 x 500. So, now, in the first layer, we will need to have 250,000 neurons. That's a huge number of neurons in the first layer for an image of this size. Hence, the network becomes computationally too expensive for the task. So, how do we solve this problem? Again, a biological inspiration comes to the rescue! Let's look at the details about the evolution of CNNs in the next section.