This chapter presents state-of-the-art deep networks for image classification.
Residual networks have become the latest architecture, with a huge improvement in accuracy and greater simplicity.
Before residual networks, there had been a long history of architectures, such as AlexNet, VGG, Inception (GoogLeNet), Inception v2,v3, and v4. Researchers were searching for different concepts and discovered some underlying rules with which to design better architectures.
This chapter will address the following topics:
Main datasets for image classification evaluation
Network architectures for image classification
Batch normalization
Global average pooling
Residual connections
Stochastic depth
Dense connections
Multi-GPU
Data augmentation techniques