Notable
VGGNet: Very Deep Convolutional Networks
VGGNet demonstrated that network depth is crucial for visual recognition. Using very small 3×3 convolutions throughout, the 16-19 layer networks achieved top performance on ImageNet while being conceptually simpler than previous architectures.
- 16-19 layers (vs. 8 in AlexNet)
- Uniform 3×3 convolution filters
- Simplicity over complexity
- Showed depth matters for CNNs
- Architecture widely adopted