人工智能培训

搜索

深度学习论文:用于单标签和多标签图像检索统一视图的层次结构邻域判别哈希(Hierarchy Neighborhood Discriminative Hashin

[复制链接]
aarryy7 发表于 2019-1-11 11:21:00 | 显示全部楼层 |阅读模式
aarryy7 2019-1-11 11:21:00 485 0 显示全部楼层
深度学习论文:用于单标签和多标签图像检索统一视图的层次结构邻域判别哈希(Hierarchy Neighborhood Discriminative Hashing for An Unified View of  Single-Label and Multi-Label Image retrieval)最近,深度监督的散列方法已经成为大规模图像检索任务的流行方法。为了保留其他示例之间的语义相似性概念,它们通常利用成对监督或三重监督信息进行散列学习。然而,这些方法通常忽略语义类信息,这有助于提高哈希码的语义差异能力。在本文中,我们提出了一种新的层次邻域判别哈希方法。具体来说,我们构造了一个二分图,以建立子类特征中心和嵌入特征之间的粗略语义邻域关系。此外,利用成对监督信息构建嵌入特征之间的细微语义邻居关系。最后,我们提出了层次邻域判别散列损失,以统一单一的labeland多标记图像检索问题与单流深度神经网络体系结构。两个大型数据集的实验结果表明,所提出的方法可以胜过最先进的散列方法。
Recently, deep supervised hashing methods have become popular for large-scaleimage retrieval task.To preserve the semantic similarity notion betweenexamples, they typically utilize the pairwise supervision or the tripletsupervised information for hash learning.However, these methods usually ignorethe semantic class information which can help the improvement of the semanticdiscriminative ability of hash codes.In this paper, we propose a novelhierarchy neighborhood discriminative hashing method.Specifically, weconstruct a bipartite graph to build coarse semantic neighbourhood relationshipbetween the sub-class feature centers and the embeddings features.Moreover, weutilize the pairwise supervised information to construct the fined semanticneighbourhood relationship between embeddings features.Finally, we propose ahierarchy neighborhood discriminative hashing loss to unify the single-labeland multilabel image retrieval problem with a one-stream deep neural networkarchitecture.Experimental results on two largescale datasets demonstrate thatthe proposed method can outperform the state-of-the-art hashing methods.深度学习论文:用于单标签和多标签图像检索统一视图的层次结构邻域判别哈希(Hierarchy Neighborhood Discriminative Hashing for An Unified View of  Single-Label and Multi-Label Image retrieval) SqGMOa81JeI1gKUG.jpg
URL地址:https://arxiv.org/abs/1901.03060     ----pdf下载地址:https://arxiv.org/pdf/1901.03060    ----深度学习论文:用于单标签和多标签图像检索统一视图的层次结构邻域判别哈希(Hierarchy Neighborhood Discriminative Hashing for An Unified View of  Single-Label and Multi-Label Image retrieval)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

aarryy7当前离线
新手上路

查看:485 | 回复:0

快速回复 返回顶部 返回列表