Facial attribute editing has mainly two objectives:1)translating image from a source domain to a target one,and 2)only changing the facial regions related to a target attribute and preserving the attribute-excluding d...Facial attribute editing has mainly two objectives:1)translating image from a source domain to a target one,and 2)only changing the facial regions related to a target attribute and preserving the attribute-excluding details.In this work,we propose a multi-attention U-Net-based generative adversarial network(MU-GAN).First,we replace a classic convolutional encoder-decoder with a symmetric U-Net-like structure in a generator,and then apply an additive attention mechanism to build attention-based U-Net connections for adaptively transferring encoder representations to complement a decoder with attribute-excluding detail and enhance attribute editing ability.Second,a self-attention(SA)mechanism is incorporated into convolutional layers for modeling long-range and multi-level dependencies across image regions.Experimental results indicate that our method is capable of balancing attribute editing ability and details preservation ability,and can decouple the correlation among attributes.It outperforms the state-of-the-art methods in terms of attribute manipulation accuracy and image quality.Our code is available at https://github.com/SuSir1996/MU-GAN.展开更多
The existing few-shot learning(FSL) approaches based on metric-learning usually lack attention to the distinction of feature contributions,and the importance of each sample is often ignored when obtaining the class re...The existing few-shot learning(FSL) approaches based on metric-learning usually lack attention to the distinction of feature contributions,and the importance of each sample is often ignored when obtaining the class representation,where the performance of the model is limited.Additionally,similarity metric method is also worthy of attention.Therefore,a few-shot learning approach called MWNet based on multi-attention fusion and weighted class representation(WCR) is proposed in this paper.Firstly,a multi-attention fusion module is introduced into the model to highlight the valuable part of the feature and reduce the interference of irrelevant content.Then,when obtaining the class representation,weight is given to each support set sample,and the weighted class representation is used to better express the class.Moreover,a mutual similarity metric method is used to obtain a more accurate similarity relationship through the mutual similarity for each representation.Experiments prove that the approach in this paper performs well in few-shot image classification,and also shows remarkable excellence and competitiveness compared with related advanced techniques.展开更多
The traditional deep learning model has problems that the longdistance dependent information cannot be learned, and the correlation between the input and output of the model is not considered. And the information proc...The traditional deep learning model has problems that the longdistance dependent information cannot be learned, and the correlation between the input and output of the model is not considered. And the information processing on the sentence set is still insufficient. Aiming at the above problems, a relation extraction method combining bidirectional GRU network and multiattention mechanism is proposed. The word-level attention mechanism was used to extract the word-level features from the sentence, and the sentence-level attention mechanism was used to focus on the characteristics of sentence sets. The experimental verification in the NYT dataset was conducted. The experimental results show that the proposed method can effectively improve the F1 value of the relationship extraction.展开更多
Marine power-generation diesel engines operate in harsh environments.Their vibration signals are highly complex and the feature information exhibits a non-linear distribution.It is difficult to extract effective featu...Marine power-generation diesel engines operate in harsh environments.Their vibration signals are highly complex and the feature information exhibits a non-linear distribution.It is difficult to extract effective feature information from the network model,resulting in low fault-diagnosis accuracy.To address this problem,we propose a fault-diagnosis method that combines the Gramian angular field(GAF)with a convolutional neural network(CNN).Firstly,the vibration signals are transformed into 2D images by taking advantage of the GAF,which preserves the temporal correlation.The raw signals can be mapped to 2D image features such as texture and color.To integrate the feature information,the images of the Gramian angular summation field(GASF)and Gramian angular difference field(GADF)are fused by the weighted average fusion method.Secondly,the channel attention mechanism and temporal attention mechanism are introduced in the CNN model to optimize the CNN learning mechanism.Introducing the concept of residuals in the attention mechanism improves the feasibility of optimization.Finally,the weighted average fused images are fed into the CNN for feature extraction and fault diagnosis.The validity of the proposed method is verified by experiments with abnormal valve clearance.The average diagnostic accuracy is 98.40%.When−20 dB≤signal-to-noise ratio(SNR)≤20 dB,the diagnostic accuracy of the proposed method is higher than 94.00%.The proposed method has superior diagnostic performance.Moreover,it has a certain anti-noise capability and variable-load adaptive capability.展开更多
基金supported in part by the National Natural Science Foundation of China(NSFC)(62076093,61871182,61302163,61401154)the Beijing Natural Science Foundation(4192055)+3 种基金the Natural Science Foundation of Hebei Province of China(F2015502062,F2016502101,F2017502016)the Fundamental Research Funds for the Central Universities(2020YJ006,2020MS099)the Open Project Program of the National Laboratory of Pattern Recognition(NLPR)(201900051)The authors gratefully acknowledge the support of NVIDIA Corporation with the donation of the GPU used for this research.
文摘Facial attribute editing has mainly two objectives:1)translating image from a source domain to a target one,and 2)only changing the facial regions related to a target attribute and preserving the attribute-excluding details.In this work,we propose a multi-attention U-Net-based generative adversarial network(MU-GAN).First,we replace a classic convolutional encoder-decoder with a symmetric U-Net-like structure in a generator,and then apply an additive attention mechanism to build attention-based U-Net connections for adaptively transferring encoder representations to complement a decoder with attribute-excluding detail and enhance attribute editing ability.Second,a self-attention(SA)mechanism is incorporated into convolutional layers for modeling long-range and multi-level dependencies across image regions.Experimental results indicate that our method is capable of balancing attribute editing ability and details preservation ability,and can decouple the correlation among attributes.It outperforms the state-of-the-art methods in terms of attribute manipulation accuracy and image quality.Our code is available at https://github.com/SuSir1996/MU-GAN.
基金Supported by the National Natural Science Foundation of China (No.61171131)Key R&D Program of Shandong Province (No.YD01033)。
文摘The existing few-shot learning(FSL) approaches based on metric-learning usually lack attention to the distinction of feature contributions,and the importance of each sample is often ignored when obtaining the class representation,where the performance of the model is limited.Additionally,similarity metric method is also worthy of attention.Therefore,a few-shot learning approach called MWNet based on multi-attention fusion and weighted class representation(WCR) is proposed in this paper.Firstly,a multi-attention fusion module is introduced into the model to highlight the valuable part of the feature and reduce the interference of irrelevant content.Then,when obtaining the class representation,weight is given to each support set sample,and the weighted class representation is used to better express the class.Moreover,a mutual similarity metric method is used to obtain a more accurate similarity relationship through the mutual similarity for each representation.Experiments prove that the approach in this paper performs well in few-shot image classification,and also shows remarkable excellence and competitiveness compared with related advanced techniques.
文摘The traditional deep learning model has problems that the longdistance dependent information cannot be learned, and the correlation between the input and output of the model is not considered. And the information processing on the sentence set is still insufficient. Aiming at the above problems, a relation extraction method combining bidirectional GRU network and multiattention mechanism is proposed. The word-level attention mechanism was used to extract the word-level features from the sentence, and the sentence-level attention mechanism was used to focus on the characteristics of sentence sets. The experimental verification in the NYT dataset was conducted. The experimental results show that the proposed method can effectively improve the F1 value of the relationship extraction.
基金supported by the Project of Shanghai Engineering Research Center for Intelligent Operation and Maintenance and Energy Efficiency Monitoring of Ships(No.20DZ2252300),China.
文摘Marine power-generation diesel engines operate in harsh environments.Their vibration signals are highly complex and the feature information exhibits a non-linear distribution.It is difficult to extract effective feature information from the network model,resulting in low fault-diagnosis accuracy.To address this problem,we propose a fault-diagnosis method that combines the Gramian angular field(GAF)with a convolutional neural network(CNN).Firstly,the vibration signals are transformed into 2D images by taking advantage of the GAF,which preserves the temporal correlation.The raw signals can be mapped to 2D image features such as texture and color.To integrate the feature information,the images of the Gramian angular summation field(GASF)and Gramian angular difference field(GADF)are fused by the weighted average fusion method.Secondly,the channel attention mechanism and temporal attention mechanism are introduced in the CNN model to optimize the CNN learning mechanism.Introducing the concept of residuals in the attention mechanism improves the feasibility of optimization.Finally,the weighted average fused images are fed into the CNN for feature extraction and fault diagnosis.The validity of the proposed method is verified by experiments with abnormal valve clearance.The average diagnostic accuracy is 98.40%.When−20 dB≤signal-to-noise ratio(SNR)≤20 dB,the diagnostic accuracy of the proposed method is higher than 94.00%.The proposed method has superior diagnostic performance.Moreover,it has a certain anti-noise capability and variable-load adaptive capability.