Deep Discriminative Feature Learning Method For Object Recognition

This chapter introduces two deep discriminative feature learning methods for object recognition without the need to increase the network complexity, one based on entropy-orthogonality loss, and another one based on Min-Max loss. These two losses can enforce the learned feature vectors to have better within-class compactness and between-class separability. Therefore the discriminative ability of the learned feature vectors is highly improved, which is very essential to object recognition.