Integrating information theory and adversarial learning for cross-modal retrieval

作者：

Highlights：

• Combining information theory and adversarial learning into an end-to-end framework.

• Reducing the heterogeneity gap for cross-modal retrieval by reducing the measured modality uncertainty.

• Introducing a term based on KL-divergence to address the data imbalance issue in visual and textual modality.

摘要

•Combining information theory and adversarial learning into an end-to-end framework.•Reducing the heterogeneity gap for cross-modal retrieval by reducing the measured modality uncertainty.•Introducing a term based on KL-divergence to address the data imbalance issue in visual and textual modality.

论文关键词：Cross-modal retrieval,Shannon information theory,Adversarial learning,Modality uncertainty,Data imbalance

论文评审过程：Received 20 November 2019, Revised 1 November 2020, Accepted 31 March 2021, Available online 8 April 2021, Version of Record 17 April 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.107983