人工智能培训

搜索

论文代码开源:自己动手单相机3D指针输入设备(Do-It-Yourself Single Camera 3D Pointer Input Device)

[复制链接]
admin 发表于 2018-9-15 10:09:42 | 显示全部楼层 |阅读模式
admin 2018-9-15 10:09:42 1255 0 显示全部楼层
人工智能论文代码开源:自己动手单相机3D指针输入设备(Do-It-Yourself Single Camera 3D Pointer Input Device)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch。在本文中,我们提出了一个半监督框架来解决当前正则化方法中发现的过平滑问题。我们仔细地建议通过构造相似的聚类来推导正则化方法。我们提出稀疏标签平滑正则化(SLSR),它包括三个步骤。首先,我们训练CNN学习标记数据的判别模式。对于每个图像,我们从最后一个卷积层中提取特征图,并直接在特征上应用\ textit {k-means}聚类算法。其次,我们训练GAN模型进行特征表示学习,并为每个聚类生成样本图像。使用我们的正则化方法为每个生成的样本分配标签。第三,我们定义了一个新的目标函数和微调的两个基线模型ResNet和DenseNet。四个大型数据集Market-1501,CUHK03,DukeMTMC-ReID和VIPeR的广泛实验表明,与现有的半监督方法相比,我们的正则化方法显着提高了Re-IDaccuracy。例如,在Market-1501数据集上,ResNet的秩-1准确度从87.29%提高到89.16%,DenseNet从90.05%提高到92.43%。该代码可通过此https URL获得
In this paper, we propose a semi-supervised framework to address theover-smoothness problem found in current regularization methods.We carefullypropose to derive a regularization method by constructing clusters of similarimages.We propose Sparse Label Smoothing Regularization (SLSR) which consistof three steps.First, we train a CNN to learn discriminative patterns fromlabeled data.For each image, we extract the feature map from the lastconvolution layer and directly apply \textit{k-means} clustering algorithm onthe feature.Secondly, we train a GAN model for feature representation learningand generate sample images for each cluster.Each generated sample is assigneda label using our regularization method.Thirdly, we define a new objectivefunction and fine-tuned two baseline models ResNet and DenseNet.Extensiveexperiments on four large-scale datasets Market-1501, CUHK03, DukeMTMC-ReID,and VIPeR show that our regularization method significantly improves the Re-IDaccuracy compared to existing semi-supervised methods.On Market-1501 dataset,for instance, rank-1 accuracy is improved from 87.29% to 89.16% for ResNet, andfrom 90.05% to 92.43% for DenseNet.The code is available atthis https URL论文代码开源:自己动手单相机3D指针输入设备(Do-It-Yourself Single Camera 3D Pointer Input Device) S4Sa4z34skKk7sN2.jpg
URL地址:https://arxiv.org/abs/1809.04976v1     ----pdf下载地址:https://arxiv.org/pdf/1809.04976v1    ----         ----github下载地址:https://github.com/jpainam/SLS_ReID    ----    论文代码开源:自己动手单相机3D指针输入设备(Do-It-Yourself Single Camera 3D Pointer Input Device)请注意该人工智能论文代码开源在github,大部分是python写的,框架可能是tensorflow或者pytorch,keras,至于具体是哪一个没有完全测试。
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

admin当前离线
管理员

查看:1255 | 回复:0

快速回复 返回顶部 返回列表