检索结果-内蒙古大学图书馆

IEEE Conference on Computer Vision and pattern recognition

作者： Jiaxiang Wu Cong Leng Yuhang Wang Qinghao Hu Jian Cheng Nat. Lab. of Pattern Recognition Inst. of Autom. Beijing China

ISBN: (纸本)9781467388528

Recently, convolutional neural networks (CNN) have demonstrated impressive performance in various computer vision tasks. However, high performance hardware is typically indispensable for the application of CNN models due to the high computation complexity, which prohibits their further extensions. In this paper, we propose an efficient framework, namely Quantized CNN, to simultaneously speed-up the computation and reduce the storage and memory overhead of CNN models. Both filter kernels in convolutional layers and weighting matrices in fully-connected layers are quantized, aiming at minimizing the estimation error of each layer's response. Extensive experiments on the ILSVRC-12 benchmark demonstrate 4 ～ 6× speed-up and 15 ～ 20× compression with merely one percentage loss of classification accuracy. With our quantized CNN model, even mobile devices can accurately classify images within one second.

关键词： Quantization (signal) Convolutional codes Kernel Computational modeling Estimation error Acceleration Image coding

来源：评论

学校读者我要写书评

暂无评论

Object Detection by lab.ling Superpixels

Object Detection by Labeling Superpixels

引用

IEEE Conference on Computer Vision and pattern recognition

作者： Junjie Yan Yinan Yu Xiangyu Zhu Zhen Lei Stan Z. Li Nat. Lab. of Pattern Recognition China

ISBN: (纸本)9781467369657

Object detection is often conducted by object proposal generation and classification sequentially. This paper handles object detection in a superpixel oriented manner instead of the proposal oriented. Specially, this paper takes object detection as a multi-lab.l superpixel lab.ling problem by minimizing an energy function. It uses the data cost term to capture the appearance, smooth cost term to encode the spatial context and lab.l cost term to favor compact detection. The data cost is learned through a convolutional neural network and the parameters in the lab.ling model are learned through a structural SVM. Compared with proposal generation and classification based methods, the proposed superpixel lab.ling method can nat.rally detect objects missed by proposal generation step and capture the global image context to infer the overlapping objects. The proposed method shows its advantage in Pascal VOC and ImageNet. Notably, it performs better than the ImageNet ILSVRC2014 winner GoogLeNet (45.0% V.S. 43.9% in mAP) with much shallower and fewer CNNs.

关键词： Object detection spatial context

来源：评论

学校读者我要写书评

暂无评论

Estimate articulatory MRI series from acoustic signal using deep architecture

Estimate articulatory MRI series from acoustic signal using ...

引用

IEEE Internat.onal Conference on Acoustics, Speech and Signal Processing

作者： H. Li J. Tao M. Yang B. Liu Nat. Lab. of Pattern Recognition Inst. of Autom. Beijing China

ISBN: (纸本)9781467369985

This paper presents our work on acoustic-to-articulatory inversion mapping, in which, the articulatory data is the MRI series for articulators on mid-sagittal plan. Deep architectures based on restricted Boltzmann machine (RBM) and linear regression are employed to construct the audio-visual mapping. We test two architectures to initialize the neural network: the bottom-up stacked RBM with top regression layer architecture and the one with extra Gaussian-Bernoulli RBM on the top of the former architecture. GMM-based mapping is used as baseline method. The MRI data from USC-TIMIT database is used for the training. The experimental results show that the deep regression network is an effective model to construct the mapping from acoustic speech signal to articulatory MRI series, and also indicate that it is a better strategy to initial the top layer as Gaussian-Bernoulli RBM to compress the MRI data before the liner regression.

关键词： MRI acoustic-to-articulatory inversion deep neural network deep regression network Magnetic Resonance Imaging C7orf49 gene audio signals ARCHITECTURE Voice Signal MAPPING(MATHEMATICAL) Articulators baseline method Building construction top layer series (mathematics)

来源：评论

学校读者我要写书评

暂无评论

Ordinal pyramid pooling for rotation invariant object recognition

Ordinal pyramid pooling for rotation invariant object recogn...

引用

IEEE Internat.onal Conference on Acoustics, Speech and Signal Processing

作者： G. Wang B. Fan C. Pan Nat. Lab. of Pattern Recognition Inst. of Autom. Beijing China

ISBN: (纸本)9781467369985

Local feature descriptor plays a fundamental role in many visual tasks, and its rotation invariance is a key issue for many recognition and detection problems. This paper proposes a novel rotation invariant descriptor by ordinal pyramid pooling of local Fourier transform features based on their radial gradient orientations. Since both the low-level feature and pooling strategy are rotation invariant, the obtained descriptor is rotation invariant by nat.re. Pooling based on orders of gradient orientations is not only invariant to in-plane rotation, but also encodes gradient orientation information into descriptor as well as spatial information to some extent. Moreover, these information is enhanced by the proposed pyramid pooling structure. Therefore, our method is nat.rally rotation invariant and has strong discriminat.ve ability. Experimental results on the aerial car dataset demonstrate the effectiveness of our descriptor.

关键词： Local feature descriptor Orders of radial gradient orientations Ordinal pyramid Rotation invariant pooling Descriptors Invariant Pyramid Plane rotation VISUAL TASKS Ordinal

来源：评论

学校读者我要写书评

暂无评论

Voice quality: Not only about "you" but also about "your interlocutor"

Voice quality: Not only about "you" but also about "your int...

引用

IEEE Internat.onal Conference on Acoustics, Speech and Signal Processing

作者： Y. Li N. Campbell J. Tao Nat. Lab. of Patternattern Recognition Inst. of Autom. Beijing China

ISBN: (纸本)9781467369985

This paper investigates the effect of voice quality in commutative speech. Voice quality is often considered as the characteristic auditory colouring of an individual speaker's voice, but in our study, we find that voice quality can also reveal information about the interlocutor in everyday social interactions. In the correlation analysis between acoustic measures and interlocutors, the effect caused by the linguistic content was reduced by focusing on the corpus of the commonly-used Japanese word "yes". The distributions of voice quality features, e.g., Normalized Amplitude Quotient (NAQ), jitter and shimmer, showed a clear difference among different interlocutors, e.g., friend or business partner. Automatic classification of interlocutors was conducted using random forest method with voice quality, prosodic and spectral features. The best classification accuracy was 76.9% over four interlocutors' data.

关键词： Voice quality commutative speech decision trees prosody social signal Voice Quality prosody Decision Trees Automatic classification

来源：评论

学校读者我要写书评

暂无评论

WINDOW MINING BY CLUSTERING MID-LEVEL REPRESENTATION FOR WEAKLY SUPERVISED OBJECT DETECTION

WINDOW MINING BY CLUSTERING MID-LEVEL REPRESENTATION FOR WEA...

引用

IEEE Internat.onal Conference on Image Processing

作者： Chong Wang Weiqiang Ren Kaiqi Huang Nat. Lab. of Pattern Recognition Inst. of Autom. Beijing China

ISBN: (纸本)9781479957521

Discovering positive detection windows in training images is a challenging problem in weakly supervised object detection. In this paper, we propose a window mining strategy by the simple and efficient k-means clustering. Firstly, a recent segmentation based object proposal is used for its highly semantic candidate windows;secondly, the bag-of-words model is adopted as mid-level object representation for each window. By clustering these windows with k-means, semantic clusters can be generated. Then, to discover the positive windows from these clusters, we further propose a cluster selection method based on each cluster's discriminat.on, which is evaluated by classification performance given the category lab.l. With the semantic clusters, this selection process is effective and efficient. Evaluation on the challenging PASCAL VOC 2007 dataset shows that the proposed method outperforms all previous weakly supervised approaches.

关键词： data mining image representation image segmentation learning (artificial intelligence) object detection pattern clustering PASCAL VOC 2007 dataset bag-of-words model cluster selection method detection windows discovery k-means clustering mid-level object representation mid-level representation clustering segmentation based object proposal weakly supervised object detection window mining strategy Feature extraction Motorcycles Object detection Proposals Semantics Training Visualization k-means weakly supervised learning window mining

来源：评论

学校读者我要写书评

暂无评论

A Way to Apply Traditional Clustering Methods in Bi-Cluster Detection

A Way to Apply Traditional Clustering Methods in Bi-Cluster ...

引用

Internat.onal Conference on Biomedical Engineering and Computer Science (ICBECS)

作者： Yanjie Zhang Hong Wang Zhanyi Hu Sch. of Comput. Sci. & Technol. Yantai Univ. Yantai China Inst. of Botany Chinese Acad. of Sci. China Nat. Lab. of Pattern Recognition CAS Beijing China

Bi-clustering microarry data is very important for the research on gene regulatory mechanisms. Genes which exhibit similar patterns are often functionally related. Compared with the traditional data clustering methods, bicluster detection is highly different. The former usually use the whole row or column vectors as feature vectors. While as to the latter the elements in a founded bicluster are greatly distributed among the original data matrix. Experiments have shown that when the number of the biclusters is unknown even an obvious bicluster which can be found by human beings may be wrongly clustered. In this paper a novel bicluster detection framework is proposed, which tries to make fully use of the rich and powerful existing data clustering methods. The clustering results are used for further bicluster detection. Based on an interesting phenomenon which can be proved to be a characteristic of bicluster, the biclusters are detected one by one. At the end of the paper experiment results on the simulated data are presented.

关键词： Clustering methods Gene expression DNA Computer science lab.ratories pattern recognition Automation Humans Vectors Time measurement

来源：评论

学校读者我要写书评

暂无评论

A Sequential Gene Expression Data Bi-clustering Method: Clustering and Verification

A Sequential Gene Expression Data Bi-clustering Method: Clus...

引用

Internat.onal Conference on Artificial Intelligence and Computational Intelligence (AICI)

作者： Yanjie Zhang Zhanyi Hu Sch. of Comput. Sci. & Technol. Yantai Univ. Yantai China Nat. Lab. of Pattern Recognition Inst. of Autom. Beijing China

With the development of the technique of DNA chips, more and more experiments data can be gained. Genes exhibiting similar patterns are often functionally related, which is very helpful for the research on analyzing the underlying mechanisms of metabolic and regulatory networks in the cell. But the fact that the correlated part is entangled with the unrelated part in the data, and the data is usually corrupted by some biological or experimental factors adds great difficulty to detect all of the possible biclusters. The novelty of the method proposed in this paper lies in the given data is not treated as a unique one but is segmented into many 2-combinat.ons. And clustering is done with all these 2-combinat.ons. Then based on the clustering results some binary tables are created. The second part is to verify the concatenat.d quasi-biclusters. The whole processing aims at finding all of the possible biclusters from large to small. At the end of the paper, an experiment result with a simulated small but complicated data is presented.

关键词： Gene expression DNA Data analysis Artificial intelligence Computational intelligence Computer science pattern analysis Concatenat.d codes lab.ratories pattern recognition

来源：评论

学校读者我要写书评

暂无评论

Prosodic word prediction using the lexical information

Prosodic word prediction using the lexical information

引用

IEEE Internat.onal Conference on nat.ral Language Processing and Knowledge Engineering (NLP-KE)

作者： Honghui Dong JianhuaTao Bo Xu National Laboratory of Pattern Recognition China Nat. Lab. of Pattern Recognition Chinese Acad. of Sci. Beijing China High Technology Innovation Center Institute of Automation Chinese Academy of Science China

As a basic prosodic unit, the prosodic word influences the nat.ralness and the intelligibility greatly. Although the relation between the lexicon word and the prosodic word has been widely discussed, the prosodic word prediction still cannot reach a high precision and recall. In this paper, the study shows the lexical features are more efficient in prosodic word prediction. Based on careful analysis on the mapping relationship and the difference between the lexicon words and the prosodic words, this paper proposes two methods to predict prosodic words from the lexicon words sequence. The first is a statistic probability model, which efficiently combines the local POS and word length information. Experiments show that by choosing appropriate threshold this statistic model can reach a high precision and high recall ratio. Another is an SVM-based method. In this SVM classifier, an efficient feature is introduced. Besides the POS and the word length features, the in-word-probability (IWP) is used. IWP means the probability of a lexicon word to be in a prosodic word. The precision and the recall ratio are improved greatly after using SVM classifier and the IWP feature.

关键词： Support vector machines Support vector machine classification Statistics Probability Speech synthesis Predictive models lab.ratories pattern recognition Technological innovation Automation

来源：评论

学校读者我要写书评

暂无评论

A NON-PARAMETER BAYESIAN CLASSIFIER FOR FACE recognition

引用

Journal of Electronics(China) 2003年第5期20卷 362-370页

作者： Liu Qingshan Lu Hanqing Ma Songde (nat. lab.of pattern recognition, Inst. of Automation, Chinese Academy of Sciences, Beijing 100080) Nat. Lab of Pattern Recognition Inst. of Automation Chinese Academy of Sciences Beijing

A non-parameter Bayesian classifier based on Kernel Density Estimation (KDE)is presented for face recognition, which can be regarded as a weighted Nearest Neighbor (NN)classifier in formation. The class conditional density is estimated by KDE and the bandwidthof the kernel function is estimated by Expectation Maximum (EM) algorithm. Two subspaceanalysis methods-linear Principal Component Analysis (PCA) and Kernel-based PCA (KPCA)are respectively used to extract features, and the proposed method is compared with ProbabilisticReasoning Models (PRM), Nearest Center (NC) and NN classifiers which are widely used in facerecognition systems. The experiments are performed on two benchmarks and the experimentalresults show that the KDE outperforms PRM, NC and NN classifiers.

关键词： Kernel Density Estimation (KDE) Probabilistic Reasoning Models (PRM) Principal Component Analysis (PCA) Kernel-based PCA (KPCA) Face recognition

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：