This website requires JavaScript.

Asymmetric Transfer Hashing with Adaptive Bipartite Graph Learning

Jianglin LuJie ZhouYudong ChenWitold PedryczZhihui LaiKwok-Wai Hung
Jun 2022
摘要
Thanks to the efficient retrieval speed and low storage consumption, learningto hash has been widely used in visual retrieval tasks. However, existinghashing methods assume that the query and retrieval samples lie in homogeneousfeature space within the same domain. As a result, they cannot be directlyapplied to heterogeneous cross-domain retrieval. In this paper, we propose aGeneralized Image Transfer Retrieval (GITR) problem, which encounters twocrucial bottlenecks: 1) the query and retrieval samples may come from differentdomains, leading to an inevitable {domain distribution gap}; 2) the features ofthe two domains may be heterogeneous or misaligned, bringing up an additional{feature gap}. To address the GITR problem, we propose an Asymmetric TransferHashing (ATH) framework with its unsupervised/semi-supervised/supervisedrealizations. Specifically, ATH characterizes the domain distribution gap bythe discrepancy between two asymmetric hash functions, and minimizes thefeature gap with the help of a novel adaptive bipartite graph constructed oncross-domain data. By jointly optimizing asymmetric hash functions and thebipartite graph, not only can knowledge transfer be achieved but informationloss caused by feature alignment can also be avoided. Meanwhile, to alleviatenegative transfer, the intrinsic geometrical structure of single-domain data ispreserved by involving a domain affinity graph. Extensive experiments on bothsingle-domain and cross-domain benchmarks under different GITR subtasksindicate the superiority of our ATH method in comparison with thestate-of-the-art hashing methods.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答