Weighted Contrastive Hashing

The development of unsupervised hashing is advanced by the recent popular contrastive learning paradigm. However, previous learning-based works have been hampered (1) insufficient data similarity mining based on global-only image representations, and (2) hash code semantic loss caused augmentation. In this paper, we propose a novel method, namely Weighted Contrative Hashing (WCH), to take step towards solving these two problems. We introduce mutual attention module alleviate problem information asymmetry in network features missing structure during contrative Furthermore, explore fine-grained relations between images, i.e., divide images into multiple patches calculate similarities patches. aggregated weighted similarities, which reflect deep relations, are distilled facilitate codes with distillation loss, so as obtain better retrieval performance. Extensive experiments show that proposed WCH significantly outperforms existing methods three benchmark datasets. Code available at: http://github.com/RosieYuu/WCH .