Study
Computer Vision · CNN architectures: VGG and ResNet
VGG's stacks of 3×3 convolutions, why very deep plain networks got worse, and ResNet's skip connections, with a residual block checked in code.
- Two stacked 3×3 convolutions see a 5×5 region with fewer parameters than one 5×5 convolution.
- A residual block learns F(x) and outputs F(x) + x, so an extra layer can at worst learn to do nothing.