SO-softmax loss for discriminable embedding learning in CNNs. (November 2022)

• Record Type: Journal Article
• Authors: Zhang, Qiang; Yang, Jibin; Zhang, Xiongwei; Cao, Tieyong
• Is Part Of: Pattern recognition. Volume 131(2022)
• Publication Date: 2022-11
• Subjects: Convolutional neural networks -- Cosine similarity -- Cross entropy loss -- Quadratic transformation -- Embedding learning -- Softmax; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108877
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: A generalized softmax loss deduces to various variants of softmax loss. Goal optimization-based transformation constrains inter/intra-class cosine similarity. The proposed transformation unifies cosine similarity transformations used in losses. SO-softmax loss is proposed to enhance the embeddings' discriminability in CNN. Extensive experiments show the superiority of SO-softmax over other counterparts. Abstract: Convolutional neural networks (CNNs)-based classifiers, trained with the softmax cross-entropy loss, have achieved remarkable success in learning embeddings for pattern recognition. The cosine similarity-based softmax variants further improve the performance by focusing on optimizing the angles between embeddings and class weights. However, embeddings learned by these variants still have significant intra-class variances since these methods only optimize the relative differences between intra- and inter-class cosine similarities. To simultaneously optimize intra- and inter-class cosine similarities, this paper proposes a cosine Similarity Optimization-based softmax (SO-softmax) loss, which is based on a generalized softmax loss formulation that combines both similarities. The proposed loss constrains the intra-class (positive) and inter-class (negative) cosine similarity by quadratic transformations, thus making the embedding representation more compact within classes and more distinguishable between classes. It is verified theoretically that SO-softmax loss can optimize both the similarities simultaneously. Thorough experiments are conducted on typical audio classification, image classification, face verification, image retrieval, and person re-identification tasks, and the results show that SO-softmax loss outperforms the state-of-the-art loss functions in CNNs-based frameworks.