ID 原文 译文
19375 最后使用Office-31数据集的6个迁移任务和Office-Home数据集的12个迁移任务进行了实验，该方法在14个迁移任务上取得了提升，在平均精度上分别提升1.4%和3.1%。 Finally, experiments are carried out using the six domain adaptation tasks of the Office-31 dataset and the 12 domain adaptation tasks of the Office-Home dataset. The proposed method improves the 14 domain adaptation tasks and increases the averageaccuracy by 1.4% and 3.1% respectively.
19376 该文提出了一种结合区域和深度残差网络的语义分割模型。基于区域的语义分割方法使用多尺度提取相互重叠的区域，可识别多种尺度的目标并得到精细的物体分割边界。 An image semantic segmentation model based on region and deep residual network is proposed.Region based methods use multi-scale to create overlapping regions, which can identify multi-scale objects andobtain fine object segmentation boundary.
19377 基于全卷积网络的方法使用卷积神经网络(CNN)自主学习特征，可以针对逐像素分类任务进行端到端训练，但是这种方法通常会产生粗糙的分割边界。 Fully convolutional methods learn features automatically by using Convolutional Neural Network (CNN) to perform end-to-end training for pixel classification tasks, but typically produce coarse segmentation boundaries.
19378 该文将两种方法的优点结合起来：首先使用区域生成网络在图像中生成候选区域，然后将图像通过带扩张卷积的深度残差网络进行特征提取得到特征图，结合候选区域以及特征图得到区域的特征，并将其映射到区域中每个像素上； The advantages of these two methods are combined: firstly, candidateregions are generated by region generation network, and then the image is fed through the deep residualnetwork with dilated convolution to obtain the feature map. Then the candidate regions and the feature mapsare combined to get the features of the regions, and the features are mapped to each pixel in the regions.
19379 最后使用全局平均池化层进行逐像素分类。 Finally, the global average pooling layer is used to classify pixels.
19380 该文还使用了多模型融合的方法，在相同的网络模型中设置不同的输入进行训练得到多个模型，然后在分类层进行特征融合，得到最终的分割结果。 Multiple different models are obtained bytraining with different sizes of candidate region inputs. When testing, the final segmentation are obtained by fusing the classification results of these models.
19381 在SIFT FLOW和PASCALContext数据集上的实验结果表明该文方法具有较高的平均准确率。 The experimental results on SIFT FLOW and PASCALContext datasets show that the proposed method has higher average accuracy than some state-of-the-artalgorithms.
19382 为优化软件定义网络(SDN)的路由选路，该文将深度增强学习原理引入到软件定义网络的选路过程，提出一种基于深度增强学习的路由优化选路机制，用以削减网络运行时延、提高吞吐量等网络性能，实现连续时间上的黑盒优化，减少网络运维成本。 In order to achieve routing optimization in the Software Defined Network (SDN) environment, deep reinforcement learning is imposed to the SDN routing process and a mechanism based on deep reinforcement learning is proposed to optimize routing. This mechanism can improve network performance such as delay, throughput, and realize black-box optimization in continuous time, which surely reduces network operation and maintenance costs.
19383 此外，该文通过实验对所提出的路由优化机制进行评估，实验结果表明，路由优化机制具有良好的收敛性与有效性，较传统路由协议可提供更优的路由方案与实现更稳定的性能。 Besides, the proposed routing optimization mechanism is evaluated through a series ofexperiments. The experimental results show that the proposed SDN routing optimization mechanism has good convergence and effectiveness, and can provide better routing configurations and performance stability than traditional routing protocols.
19384 针对蜂窝网络传输性能及基站(BS)缓存能力受限，多用户内容请求难以满足用户服务质量(QoS)需求等问题，该文提出一种蜂窝终端直通(D2D)通信联合用户关联及内容部署算法。 Due to the limited transmission performance of cellular network and the buffering capabilities of theBase Station (BS), it is very difficult to achieve the Quality of Service (QoS) requirements of multi-user content requests. In this paper, a joint user association and content deployment algorithm is proposed for cellular Device-to-Device (D2D) communication network.