After the success of AlexNet in 2012, more and more researchers worked on improving the CNN architecture of AlexNet to improve accuracy. The focus shifted to smaller window size, smaller filters, and smaller strides. VGG16 was introduced in 2014 by Karen Simonyan and Andrew Zisserman in the paper titled Very Deep Convolutional Networks for Large-Scale Image Recognition. The paper can be read at https://arxiv.org/abs/1409.1556.
The model achieved a 92.7% top-five test accuracy in ImageNet in ILSVRC-2014.
The VGG16 architecture is shown in the following image:
The basic idea of VGG16 is summarized as follows:
- The maximum filter size is 3 x 3 and the minimum size is 1 x 1. This means that a smaller filter size with a larger quantity is used, compared to a larger filter size and smaller quantity for AlexNet; this results in fewer parameters compared to AlexNet...
