SO-softmax loss for discriminable embedding learning in CNNs
作者：
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.
摘要
•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.
论文关键词：Convolutional neural networks,Cosine similarity,Cross entropy loss,Quadratic transformation,Embedding learning,Softmax
论文评审过程：Received 15 July 2021, Revised 19 May 2022, Accepted 26 June 2022, Available online 27 June 2022, Version of Record 12 July 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108877
