检索结果-内蒙古大学图书馆

MoAFormer: Aggregating Adjacent Window Features into Local vision Transformer Using Overlapped Attention Mechanism for Volumetric Medical Segmentation 11

MoAFormer: Aggregating Adjacent Window Features into Local V...

引用

11th International Conference on Computing and pattern recognition, ICCPR 2022

作者： Luo, Yixi Yin, Huayi Du, Xia Department of Computer and Information Engineering Fujian Provincial Key Laboratory of Pattern Recognition and Image Understanding Xiamen University of Technology China

ISBN: (纸本)9781450397056

The window-based attention is used to alleviate the problem of abrupt increase in computation as the input image resolution grows and shows excellent performance. However, the problem that aggregating global features from different windows is waiting to be resolved. Swin-Transformer is proposed to construct hierarchical encoding by a shifted-window mechanism to interactively learn the information between different windows. In this work, we investigate the outcome of applying an overlapped attention block (MoA) after the local attention layer and apply plenty to medical image segmentation tasks. The overlapped attention module employs slightly larger and overlapped patches in the key and value to enable neighbouring pixel information transmission, which leads to significant performance gain. The experimental results on the ACDC and Synapse datasets demonstrate that the used method performs better than previous Transformer models. © 2022 ACM.

关键词： Image resolution

来源：评论

学校读者我要写书评

暂无评论

SIR-HCL: Semantic-Inconsistency Reasoning and Hybrid Contrastive Learning for Efficient Cross-Emotion Anomaly Detection

引用

IEEE Transactions on Cognitive and Developmental Systems 2025年

作者： Liu, Xin Chen, Qiyan Cheung, Yiu-Ming Peng, Shu-Juan Huaqiao University Department of Computer Science Xiamen361021 China Hong Kong Baptist University Department of Computer Science SAR Hong Kong Hong Kong Xiamen Key Laboratory of Computer Vision and Pattern Recognition Xiamen361021 China Huaqiao University Fujian Key Laboratory of Big Data Intelligence and Security Xiamen361021 China Huaqiao University Department of Artificial Intelligence Xiamen China Fujian Province University Key Laboratory of Computer Vision and Machine Learning Huaqiao University Xiamen361021 China

Cross-emotion anomaly detection is an emerging and challenging research topic in cognitive analysis field, which aims at identifying the abnormal emotion pair whose semantic patterns are inconsistent across different emotional modalities. To the best of our knowledge, this topic has yet to be well studied, which could potentially benefit lots of valuable cognitive applications such as autistic children diagnosis and criminal deception detection. To this end, this paper proposes an efficient cross-emotion anomaly detection approach via semanticinconsistency reasoning and hybrid contrastive learning (SIR-HCL), which is the first attempt to detect the anomalous emotional pairs across the audio-visual emotions. First, the proposed framework utilizes dual-branch network to obtain the deep emotional features in each modality, and then employs the shared residual block to derive the semantically compatible features. Subsequently, an efficient hybrid contrastive learning approach is designed to enlarge the semantic-inconsistency among abnormal emotional pair with different affective classes, while enhancing the semantic-consistency and increasing the feature correlation between normal emotional pair from the same affective class. At the same time, an efficient bidirectional learning scheme is employed to significantly improve the data utilization and a two-component Beta Mixture Model is adaptively utilized to reason the anomalous emotion pairs. Extensive experiments evaluated on two benchmark datasets show that the proposed SIR-HCL method can well detect the anomalous emotional pairs across audio-visual emotional data, and brings substantial improvements over the state-of-the-art competing methods. © 2016 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

A Dynamic Spatio-temporal Network with Self-attention for Multi-station Passenger Flow Prediction

A Dynamic Spatio-temporal Network with Self-attention for Mu...

引用

International Conference on Awareness Science and Technology (iCAST)

作者： Fengzhi Wang Qinzhi Lv Lijuan Liu College of Computer and Information Engineering Xiamen University of Technology Xiamen China Fujian Key Laboratory of Pattern Recognition and Image Understanding

Passenger flow prediction is vitally significant for intelligent transportation systems (ITS). Most of the studies typically focus on the passenger flow prediction for an individual station, and only capture the temporal features without considering any spatial features. Constructing a passenger flow prediction model for multiple stations, or even a whole network, is more valuable for practical applications. Therefore, we develop a dynamic spatio-temporal network (DSTNet) with a self-attention (SA) mechanism for multi-station passenger flow prediction. A dynamic graph convolutional network (DGCN) is applied for the spatial feature extraction, and gated recurrent unit (GRU) is combined to learn the temporal features. SA is applied to further assign the weights for the extracted spatio-temporal features. The Experiment has been conducted on the passenger flow in the xiamen bus rapid transit (BRT). The results demonstrate that the proposed DSTNet with SA (SA-DSTNet) outperforms the baselines in the multi-station passenger flow prediction task.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Interactive Semantic Segmentation With Weak Supervision 22

Interactive Semantic Segmentation With Weak Supervision

引用

8th International Conference on Computing and Artificial Intelligence, ICCAI 2022

作者： Gong, Lei Wang, Da-Han Wu, Yun Ye, Hai-Li Zhu, Chen-Yan School of Computer and Information Engineering Xiamen University of Technology Xiamen361024 China Fujian Key Laboratory of Pattern Recognition and Image Understanding Xiamen361024 China Medical Diagnostic Systems Co. Ltd. Xiamen361000 China

ISBN: (纸本)9781450396110

At present, the most advanced semantic segmentation model training mainly relies on pixel-level annotation, that is, annotating the category of each pixel of an image. Such annotation usually is time-consuming and expensive, especially for special applications that require expert annotation. The weakly-supervised segmentation method using the point-level supervision information has been investigated which however has great problems that the supervision information is quite limited and the performance is far from fully supervised methods. In this paper, we proposes an novelty interactive image segmentation method based on weak supervision, which allows multiple feedbacks of easily obtained weakly supervised information and improves the efficiency of utility of the supervision information. In the downstream task (interactive image segmentation), supervised information at the point level is used for many times, which makes the connection between pixels in the upstream task become closer and improves the segmentation accuracy. First, image-level tags are used to train the classification network. Then the pseudo-semantic labels are generated and put into the interactive segmentation network for training, and an almost completely supervised CNN is obtained, which further improves the performance and provides operability for human-computer interaction. The proposed method achieves promising semantic segmentation results that are close to those obtained by strongly supervised segmentation methods on the PASCAL VOC 2012 datasets. © 2022 ACM.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Pose focus transformer meet inter-part relation

引用

Expert Systems with Applications 2024年 240卷

作者： Luo, Yanmin Lin, Hongwei Huang, Wenlin Wang, Youjie Du, Jixiang Guo, Jing-Ming College of Computer Science and Technology Huaqiao University Xiamen361021 China Xiamen Key Laboratory of Computer Vision and Pattern Recognition Huaqiao University Xiamen361021 China Maynooth International Engineering College Fuzhou University Fuzhou350108 China Department of Electrical Engineering National Taiwan University of Science and Technology Taipei10607 China

Human pose estimation in crowded scenes is a challenging task. Due to overlap and occlusion, it is difficult to infer pose clues from individual keypoints. We proposed PFFormer, a new transformer-based approach that treats pose estimation as a hierarchical set prediction problem that first focuses on human windows and coarsely predicts whole-body poses globally within them. In PFFormer, we designed a Windows Clustering Transformer (WCT), which reorganizes the image windows by filtering the attentive windows and fusing the inattentive ones, allowing the transformer to focus on the important regions while reducing the interference from the complex background, followed by compensating for the loss of information with a global transformer. Then we partition the learned body pose into a set of structural parts and perform the Inter-Part Relation Module (IPRM) to capture the correlation between multiple parts. These full-body poses and component features are refined at a finer level through the Part-to-Joint Decoder (PJD). Extensive experiments show that PFFormer performs favorably against its counterpart on challenging datasets, including COCO2017, CrowdPose, and OChuman datasets. The performance of crowded scenes, in particular, demonstrates the robustness of the proposed methods to deal with occlusion. © 2023 Elsevier Ltd

关键词： Information filtering

来源：评论

学校读者我要写书评

暂无评论

Collaborative Weighting for Graph Convolutional Networks

Journal of Network Intelligence

引用

Journal of Network Intelligence 2023年第2期8卷 432-447页

作者： Chen, Yong Xie, Xiao-Zhu Weng, Wei Zhang, Shan-Dan Li, Tong College of Computer and Information Engineering Xiamen University of Technology Xiamen361024 China College of Computer and Information Engineering Xiamen University of Technology Fujian Key Laboratory of Pattern Recognition and Image Understanding Xiamen361024 China

Graph neural network (GNN), as a powerful method for graph representation, has attracted extensive research interest. Recently, Graph Convolutional Network (GCN) and Graph Attention Network (GAT) have shown superior performance on graph node classification and are considered as the most promising frameworks. Both GAT and GCN use a weighting mechanism in the information aggregation process. However, GCN is weighted by topological structure, while the node content is still unemployed. GAT is weighted by node content, while the topological structure remains being ignored. These weighting mechanism, which does not make full use of topology and node content, is not conducive to graph representation. In this paper, we propose a "Collaborative Weighting For Graph Convolutional Network" (CWGCN) that combines the advantages of GCN and GAT. Specifically, its weighting mechanism not only considers the topology of the graph, but also the node content. With this weighting method for representation learning, CWGCN achieves encouraging performance on a number of citation datasets in node classification. © 2023, Taiwan Ubiquitous Information CO LTD. All rights reserved.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

Efficient Image Super-Resolution Using Vast-Receptive-Field Attention 17th

Efficient Image Super-Resolution Using Vast-Receptive-Field ...

引用

17th European Conference on computer vision, ECCV 2022

作者： Zhou, Lin Cai, Haoming Gu, Jinjin Li, Zheyuan Liu, Yingqi Chen, Xiangyu Qiao, Yu Dong, Chao ShenZhen Key Lab of Computer Vision and Pattern Recognition SIAT-SenseTime Joint Lab Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Shenzhen China Shanghai AI Laboratory Shanghai China The University of Sydney Sydney Australia University of Macau Zhuhai China

ISBN: (纸本)9783031250620

The attention mechanism plays a pivotal role in designing advanced super-resolution (SR) networks. In this work, we design an efficient SR network by improving the attention mechanism. We start from a simple pixel attention module and gradually modify it to achieve better super-resolution performance with reduced parameters. The specific approaches include: (1) increasing the receptive field of the attention branch, (2) replacing large dense convolution kernels with depthwise separable convolutions, and (3) introducing pixel normalization. These approaches paint a clear evolutionary roadmap for the design of attention mechanisms. Based on these observations, we propose VapSR, the Vast-receptive-field Pixel attention network. Experiments demonstrate the superior performance of VapSR. VapSR outperforms the present lightweight networks with even fewer parameters. And the light version of VapSR can use only 21.68% and 28.18% parameters of IMDB and RFDN to achieve similar performances to those networks. The code and models are available at https://***/zhoumumu/VapSR. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

NLFA: A Non Local Fusion Alignment Module for Multi-Scale Feature in Object Detection 3rd

NLFA: A Non Local Fusion Alignment Module for Multi-Scale Fe...

引用

3rd International Symposium on Automation, Mechanical and Design Engineering, SAMDE 2022

作者： Xue, Honghui Ma, Jinshan Cai, Zheyi Fu, Junfang Guo, Feng Weng, Wei Dong, Yunxin Zhang, Zhenchang College of Computer and Information Sciences Fujian Agriculture and Forestry University Fuzhou China Fujian Zhongke Zhongxin Intelligent Technology Co. Ltd Fuzhou China Fujian Newland Auto-ID Tech. Co. Ltd Fuzhou China Department of Computer and Information Engineering Xiamen University of Technology Xiamen China Fujian Key Laboratory of Pattern Recognition and Image Understanding Xiamen China

ISBN: (纸本)9783031400698

Recently, in order to pursue better detection results, more convolutional layers and deeper networks are a direction pursued by everyone. However, more and more down-sampling convolution or up-sampling operations generate feature maps of different scales, which makes it difficult to avoid the loss of detailed information of the image, and the distribution of different scales features will be misaligned. In particular, the loss and dislocation of the target boundary information will affect the features learned by the model and reduce the accuracy. This paper proposes a feature alignment method based on non-local idea, and designed two modules—Non Local Align Module (NLA) and Channel Fusion Augment Module (CFA). At the same time, the neighborhood calculation algorithm is also designed for it, which strengthens the binding force on the calculation of boundary information. These two modules can be easily embedded into the current mainstream object detection network to improve the detection effect of the model. Compared to the previous model, the network using our NLA module and CFA module achieves better results than the original model. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Handwritten mathematical expression recognition with self-attention 4

Handwritten mathematical expression recognition with self-at...

引用

4th International Conference on Algorithms, Computing and Artificial Intelligence, ACAI 2021

作者： Chi, Xueke Wang, Da-Han Wu, Yuefeng Wu, Yun Fujian Key Laboratory of Pattern Recognition and Image Understanding Xiamen361024 China School of Computer and Information Engineering Xiamen University of Technology Xiamen361024 China

ISBN: (纸本)9781450385053

Attention-based encoder-decoder models have made great success on handwritten mathematical expression recognition in recent years. However, this kind of method has the problem of attention drift, because under the local attention mechanism based on RNN, the high similarity between coding features can cause attention confusion. To settle this problem, we propose an encoder-decoder model with self-attention, which captures the global information of the feature map and fuses the local information of the CNN as complementary features. Experiments are conducted on the CROHME2014 and CROHME 2016 competition datasets. The experimental results show that, when only using the official training dataset, the proposed method achieves recognition accuracies of 51.98% and 50.74% on the CROHME2014 and CROHME2016 competition datasets, respectively, which outperforms the other methods significantly. The improvements demonstrate the effectiveness of the self-attention module. © 2021 ACM.

关键词： Signal encoding

来源：评论

学校读者我要写书评

暂无评论

UDC-UNet: Under-Display Camera Image Restoration via U-shape Dynamic Network 17th

UDC-UNet: Under-Display Camera Image Restoration via U-shap...

引用

17th European Conference on computer vision, ECCV 2022

作者： Liu, Xina Hu, Jinfan Chen, Xiangyu Dong, Chao Shenzhen Key Lab of Computer Vision and Pattern Recognition SIAT-SenseTime Joint Lab Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Beijing China University of Macau Zhuhai China Shanghai AI Laboratory Shanghai China

ISBN: (纸本)9783031250712

Under-Display Camera (UDC) has been widely exploited to help smartphones realize full-screen displays. However, as the screen could inevitably affect the light propagation process, the images captured by the UDC system usually contain flare, haze, blur, and noise. Particularly, flare and blur in UDC images could severely deteriorate the user experience in high dynamic range (HDR) scenes. In this paper, we propose a new deep model, namely UDC-UNet, to address the UDC image restoration problem with an estimated PSF in HDR scenes. Our network consists of three parts, including a U-shape base network to utilize multi-scale information, a condition branch to perform spatially variant modulation, and a kernel branch to leverage the prior knowledge of the PSF. According to the characteristics of HDR data, we additionally design a tone mapping loss to stabilize network optimization and achieve better visual quality. Experimental results show that the proposed UDC-UNet outperforms the state-of-the-art methods in quantitative and qualitative comparisons. Our approach won second place in the UDC image restoration track of the MIPI challenge. Codes and models are available at https://***/J-FHu/UDCUNet. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Image reconstruction

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：