feature-maps

Understand why fully-connected layers fail at image scale, then build up the CNN toolkit: convolutions, kernels, stride, padding, feature maps, pooling, and parameter sharing. Finish with the ResNet residual connection idea that unlocked networks of 100+ layers.