Deep robust multilevel semantic hashing for multi-label cross-modal retrieval

作者：

Highlights：

• Keeping the minimum distance between dissimilar learned codes improves performance.

• The effective minimum distance between dissimilar codes is derived by coding theory.

• More complicated similarity structures are explored by manipulating codes.

• Triplet loss with margin-adaptive captures the fine cross-modal similarity structure.

摘要

•Keeping the minimum distance between dissimilar learned codes improves performance.•The effective minimum distance between dissimilar codes is derived by coding theory.•More complicated similarity structures are explored by manipulating codes.•Triplet loss with margin-adaptive captures the fine cross-modal similarity structure.

论文关键词：Hashing,Multi-label,Cross-modal retrieval,Deep learning

论文评审过程：Received 8 October 2020, Revised 24 May 2021, Accepted 29 May 2021, Available online 8 June 2021, Version of Record 1 July 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108084