An effective hashing method using W-Shaped contrastive loss for imbalanced datasets

作者：

Highlights：

• W-SCL minimizes the intraclass variation and maximizes the interclass variation.

• The first method for lesion retrieval with high performance and end-to-end manner.

• It is easily adaptable to changes in the number of classes and different datasets.

• It is suitable for the output of all CNN architectures.

• It generates hash codes that are discrete enough to deal with SOTAs.

摘要

•W-SCL minimizes the intraclass variation and maximizes the interclass variation.•The first method for lesion retrieval with high performance and end-to-end manner.•It is easily adaptable to changes in the number of classes and different datasets.•It is suitable for the output of all CNN architectures.•It generates hash codes that are discrete enough to deal with SOTAs.

论文关键词：CNN,Contrastive loss,Hashing,Retrieval,Skin lesion

论文评审过程：Received 15 November 2021, Revised 12 May 2022, Accepted 15 May 2022, Available online 19 May 2022, Version of Record 23 May 2022.

论文官网地址：https://doi.org/10.1016/j.eswa.2022.117612