Re-ranking image-text matching by adaptive metric fusion
作者：
Highlights：
• The uni-modal re-ranking methods are not suitable for the image-text matching.
• Fusing different metrics can provide a comprehensive similarity evaluation.
• Considering two directions in image-text matching improves re-ranking performance.
• Alleviating differences of visual and textual feature spaces improves performance.
• Our re-ranking method outperforms other methods on two representative benchmarks.
摘要
•The uni-modal re-ranking methods are not suitable for the image-text matching.•Fusing different metrics can provide a comprehensive similarity evaluation.•Considering two directions in image-text matching improves re-ranking performance.•Alleviating differences of visual and textual feature spaces improves performance.•Our re-ranking method outperforms other methods on two representative benchmarks.
论文关键词：Image-text matching,Re-ranking method,Adaptive metric fusion
论文评审过程：Received 15 February 2019, Revised 20 February 2020, Accepted 27 March 2020, Available online 31 March 2020, Version of Record 11 May 2020.
论文官网地址：https://doi.org/10.1016/j.patcog.2020.107351
