Skip to content

Very Deep Convolutional Networks for Large-Scale Image Recognition (VGGNet)

Authors: Karen Simonyan, Andrew Zisserman Year: 2015 (arXiv 2014) Citations: 50,000+

Summary

Systematic study of network depth with uniform 3x3 convolutions. VGG-16/19 demonstrate importance of depth for visual recognition accuracy.

Key Concepts

  • Small Convolutional Filters: Stacking 3x3 instead of large kernels
  • Depth Over Width: Multiple small layers vs single large layer
  • Uniform Architecture: Repeating 3x3 blocks and pooling
  • Feature Visualization: Layer-wise visualization interpretation
  • Transfer Learning: VGG features transfer effectively
  • Depth Importance: Shows crucial role of network depth

Impact

  • 50,000+ citations
  • Demonstrated critical importance of depth
  • Influenced ResNet and modern architectures
  • Pre-trained VGG models standard baseline
  • Simple, clean architecture design
  • Foundation for understanding depth effects
  • AlexNet (Krizhevsky et al., 2012)
  • ResNet (He et al., 2015)
  • Inception Networks (Szegedy et al., 2014)