【摘 要】
:
Caricature face recognition is an interesting but also a hard task for studying owing to the huge exaggeration between two quite different face modalities,photos and caricatures.So,we propose a new re
【出 处】
:
第六届中国计算机学会大数据学术会议
论文部分内容阅读
Caricature face recognition is an interesting but also a hard task for studying owing to the huge exaggeration between two quite different face modalities,photos and caricatures.So,we propose a new representation for recognition which is fused by the representation learned from photos,caricatures and generated faces.Each generated face contains 4 main facial components.Photos,caricatures and generated faces are sent to Photo ResNet,Caricature ResNet and Generated Face ResNet to learn specific representations.Then the learned three presentation are sent a fully connected layer.We adopt softmax loss and centerloss for training which can reduce the distance of intra-class.To test the performance of our proposed representation,we build a new dataset for caricature face recognition,which consists of 259 subjects,with 6490 caricatures and 8143 photos.The dataset we build is the biggest available caricature dataset.Several basic methods are used for caricature face recognition.To test the discrimination of our proposed representation,two more groups of experiments are fulfilled,including searching photos according to the selected caricature(CTP)and searching caricatures according to the selected photo(PTC),and our proposed method performs better than other convolutional neural network based representations.
其他文献
智能电网系统一直以来就是智慧城市中的紧要一环.通过对智能电网系统进行分析,可以创造出更加便利的用电服务.负荷曲线聚类是智能电网分析中的基础一环,大量后续的应用如负荷预测,用户画像构建都可以在负荷曲线聚类的基础上得以优化.本文结合最新的语音处理模型,提出一种基于卷积循环神经网络和快速傅里叶变换的方法去提取电网负荷聚类特征,同时借助三元组损失函数使该方法可以仅依靠部分标签类型的数据进行训练,进而对未出
在图数据库中,现有的基于图模拟的匹配问题主要集中在静态图的图模拟上,但是,现实生活中的许多场景,如社交网络、交通系统网络等,需要采用带有时间变化标签的时序图进行建模,因此在时序图中解决图模拟问题是必要的.由于时序图中包含的信息量相较于静态图更为庞大,并且结构更为复杂,使现有的静态图中的图模拟方法不能直接适用于时序图中.为此本文首次提出时序图的图模拟匹配定义——时序边界模拟.首先,进行模式图分割,将
随着智能电视的普及,节目付费成为电视生产企业或视频内容企业最重要的利润来源之一.挖掘潜在付费用户,促使用户付费购买增值服务越来越成为企业亟待解决的问题.本文首先基于国内最大的电视厂家之一的海信公司日志数据的特点,提出了对日志信息扩展、特征衍生以及特征提取的解决方案.结合深度模型在高阶抽象特征学习上的优势以及线性模型在低阶特征学习上的优势,提出了Simplified Wide&Deep(SWD)模型
Weather classification is getting more and more attractive because it has many potential applications,such as visual systems and intelligent transportation,especially in transportation.However,the res
蛋白质二级结构预测是生物信息学上的一个关键问题.近年来,由于深度学习的成功,本文将深度学习应用到这一问题上面,设计了一种多方面的自注意力机制的深度卷积循环网络(Multi-Aspect Self-Attentive Network,MASAN)来进行蛋白质二级结构的预测.首先,本文使用了CNN来处理氨基酸序列,提取氨基酸序列的局部特征;在此基础上,利用双向循环神经网络(Bi-GRU)处理整个氨基酸
Big data computing and analysis can uncover hidden patterns,correlations and other insights by examining large amounts of data.Comparing with the traditional processor,the new types of processors,just
为了高效地从海量的水文传感器数据中检测出异常值,提出了一种基于SparkR的水文时间序列异常检测方法.对数据进行清洗后,采用滑动窗口配合自回归积分滑动平均模型在SparkR平台上进行预测,然后对预测的结果计算置信区间,在区间范围以外的,将其判定为异常值.基于检测结果,利用K均值算法对原数据进行聚类,同时计算其状态转移概率,对检测出的异常值进行质量评估.以在滁河获取的水文传感器数据为实验数据,分别在
Research on pollution localization using sensor monitoring networks has important significance for environmental protection.There are some challenges in the detection and localization of water polluti
In the real-world many complex systems exist in the form of heterogeneous networks.As we all know,heterogeneous networks consist of various types of vertices and relations,so it is difficult to deal d
基于语义抽取的机器阅读理解是目前人工智能与大数据相结合的热点应用之一。针对复杂多文本机器阅读理解任务中的语义理解与答案提取问题,提出一种结合外部知识的动态多层次语义理解与答案抽取模型。首先,利用改进的门控单元循环神经网络匹配文本内容与问题集;然后,分别在向量化文本内容及问题集上实施多维度动态双向注意力机制分析,提高语义匹配精度;接着,利用动态指针网络确定问题答案范围,改进网络模型语义匹配效率,降低