Paper

• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Authors: Karen Simonyan, Andrew Zisserman

VGGNet

• VGGNet are series of Convolutional Neural Network models proposed by the Visual Geometry Group of Oxford University, including the well known VGG11,VGG13,VGG16 and VGG19.
• Simplicity of Architecture: VGGNet’s architecture is highly uniform, utilizing 3x3 convolutional layers stacked on top of each other..
• Increased Depth: One of the primary innovations of VGGNet was to demonstrate that the depth of the network is a critical component for achieving high performance.
• Most of the aspects for network remains same as compared to AlexNet. It uses Max-pooling layer to reduce the spatial size of feature maps
• VGGNet uses ReLU activation throughout the network.
• VGGNet was trained using a multi-scale approach where the scale of the input images was varied during the training process. This helped the model to become robust to different image sizes and resolutions.