检索结果-内蒙古大学图书馆

arXiv 2019年

作者： Chu, Tianshu Luo, Qin Yang, Jie Huang, Xiaolin Institute of Image Processing and Patter Recognition Shanghai Jiao Tong University Shanghai200240 China

Efficient model inference is an important and practical issue in the deployment of deep neural network on resource constraint platforms. Network quantization addresses this problem effectively by leveraging low-bit representation and arithmetic that could be conducted on dedicated embedded systems. In the previous works, the parameter bitwidth is set homogeneously and there is a trade-off between superior performance and aggressive compression. Actually the stacked network layers, which are generally regarded as hierarchical feature extractors, contribute diversely to the overall performance. For a well-trained neural network, the feature distributions of different categories differentiate gradually as the network propagates forward. Hence the capability requirement on the subsequent feature extractors is reduced. It indicates that the neurons in posterior layers could be assigned with lower bitwidth for quantized neural networks. Based on this observation, a simple but effective mixed-precision quantized neural network with progressively decreasing bitwidth is proposed to improve the trade-off between accuracy and compression. Extensive experiments on typical network architectures and benchmark datasets demonstrate that the proposed method could achieve better or comparable results while reducing the memory space for quantized parameters by more than 30% in comparison with the homogeneous counterparts. In addition, the results also demonstrate that the higher-precision bottom layers could boost the 1-bit network performance appreciably due to a better preservation of the original image information while the lower-precision posterior layers contribute to the regularization of k−bit networks. Copyright © 2019, The Authors. All rights reserved.

关键词： Network layers

来源：评论

学校读者我要写书评

暂无评论

Background subtraction in mobile cameras by MRF-MAP based optical flows

Background subtraction in mobile cameras by MRF-MAP based op...

引用

2013 7th International Conference on image and Graphics, ICIG 2013

作者： Zhu, Limin Zhou, Yue Institute of Image Processing and Patter Recognition Shanghai Jiao Tong University Shanghai China

ISBN: (纸本)9780769550503

Background subtraction has always been an important field in computer vision since it is generally the first step in video or image processing. It is an even more challenging job when sequences are taken by moving cameras. Unlike static surveillance cameras, the background varies a lot in every frame taken from mobile cameras. In this sense an optical flow based clustering method is proposed. The unique optical flow representing foreground movement is detected. The final output is given after a region growth in a MRF-MAP approach. Our method is evaluated on different datasets to show its effectiveness. © 2013 IEEE.

关键词： Optical flows

来源：评论

学校读者我要写书评

暂无评论

Moving object segmentation based on new likelihood functions

Moving object segmentation based on new likelihood functions

引用

18th International Symposium on Intelligent Signal processing and Communication Systems, ISPACS 2010

作者： Zhang, Xiang Liu, Zhi Yang, Jie Ministry of Education China Institute of Image Processing and Patter Recognition Shanghai Jiaotong University China

ISBN: (纸本)9781424473717

Accurate segmentation of moving objects from a video sequence is still a difficult task. A moving object segmentation method is proposed in this paper to deal with the segmentation splits and defects. First, it is claimed that the confusion point is one reason for the segmentation inaccuracy, and corresponding solution is also presented. According to the solution, new likelihood functions are proposed to compute membership probabilities, which are then used for final segmentation within an energy minimization framework. Unlike related algorithms which compute membership probabilities using kernel density estimation, the proposed method models the membership probabilities as functions with kernel density estimation as the independent variable. Experiments show that improved results are generated by the proposed likelihood functions. © 2010 IEEE.

关键词： Probability density function

来源：评论

学校读者我要写书评

暂无评论

Moving object segmentation based on new likelihood functions

Moving object segmentation based on new likelihood functions

引用

IEEE International Symposium on Intelligent Signal processing and Communication Systems (ISPACS)

作者： Xiang Zhang Zhi Liu Jie Yang Institute of Image Processing and Patter Recognition Shanghai Jiaotong University China Key Laboratory of Advanced Display and System Application Ministry of Education Shanghai University China

关键词： Radio access networks

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：