检索结果-内蒙古大学图书馆

International Conference on Networking and Network Applications (NaNA)

作者： Peng Xu Ziyi Zhao Sugang Ma Xi'an Science and Technology Museum Xi'an Shaanxi China School of Computer Science and Technology Xi'an University of Posts and Telecommunications Xi'an Shaanxi China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing School of Computer Science and Technology Xi'an University of Posts and Telecommunications Xi'an Shaanxi China

In recent years, semantic segmentation methods based on deep learning have made remarkable developments. Despite achieving high segmentation accuracy, the performance of real-time segmentation methods cannot satisfy real-world applications. In order to achieve a balance between segmentation accuracy and speed, a real-time semantic segmentation algorithm based on Tversky loss function and mixed pooling is proposed in this paper. A Short-Term Dense Concatenate network (STDC network) is used to construct the encoder, and a mixed pooling module is used for the final part of the encoder, which uses strip pooling and average pooling to enhance the feature representation while expanding the receptive field. Additionally, a Tversky-based loss function is used for the detail guidance module of the backbone network in the encoder, and a joint loss function is used to supervise the whole network's training. We achieved 76.9% mIoU at 110.7 FPS on the Cityscapes dataset, a 2.4% improvement in accuracy over the benchmark algorithm STDCSeg, and 72.2% mIoU at 177.6 FPS on the Camvid dataset, satisfied the requirements of the real-time segmentation task.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Low-Rank Reduced Biquaternion Tensor Ring Decomposition and Tensor Completion

arXiv

引用

arXiv 2025年

作者： Luo, Hui Liu, Xin Liu, Wei Zhang, Yang School of Computer Science and Engineering Faculty of Innovation Engineering Macau University of Science and Technology China Macau Institute of Systems Engineering Faculty of Innovation Engineering Macau University of Science and Technology China School of Artificial Intelligence Sun Yat-sen University The Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou China Department of Mathematics University of Manitoba MB Canada

We define the reduced biquaternion tensor ring (RBTR) decomposition and provide a detailed exposition of the corresponding algorithm RBTR-SVD. Leveraging RBTR decomposition, we propose a novel low-rank tensor completion algorithm RBTR-TV integrating RBTR ranks with total variation (TV) regularization to optimize the process. Numerical experiments on color image and video completion tasks indicate the advantages of our method. © 2025, CC BY.

关键词： Tensors

来源：评论

学校读者我要写书评

暂无评论

Boosting segmentation accuracy of the deep learning models based on the synthetic data generation 4

Boosting segmentation accuracy of the deep learning models b...

引用

4th International Workshop on Photogrammetric and Computer Vision Techniques for Video Surveillance, Biometrics and Biomedicine, PSBB 2021

作者： Danilov, V.V. Gerget, O.M. Kolpashchikov, D.Yu. Laptev, N.V. Manakov, R.A. Hérnandez-Gómez, L.A. Alvarez, F. Ledesma-Carbayo, M.J. Research Laboratory for Processing and Analysis of Big Data Tomsk Polytechnic University Tomsk Russia Department of Software Engineering Research Laboratory"Gamma Technologies" Almaty Kazakhstan Department of Signals Systems and Radiocommunications Technical University of Madrid Madrid Spain Group of Application of Visual Telecommunications Technical University of Madrid Madrid Spain Biomedical Image Technologies Group Technical University of Madrid Madrid Spain

In the era of data-driven machine learning algorithms, data represents a new oil. The application of machine learning algorithms shows they need large heterogeneous datasets that crucially are correctly labeled. However, data collection and its labeling are time-consuming and labor-intensive processes. A particular task we solve using machine learning is related to the segmentation of medical devices in echocardiographic images during minimally invasive surgery. However, the lack of data motivated us to develop an algorithm generating synthetic samples based on real datasets. The concept of this algorithm is to place a medical device (catheter) in an empty cavity of an anatomical structure, for example, in a heart chamber, and then transform it. To create random transformations of the catheter, the algorithm uses a coordinate system that uniquely identifies each point regardless of the bend and the shape of the object. It is proposed to take a cylindrical coordinate system as a basis, modifying it by replacing the Z-axis with a spline along which the h-coordinate is measured. Having used the proposed algorithm, we generated new images with the catheter inserted into different heart cavities while varying its location and shape. Afterward, we compared the results of deep neural networks trained on the datasets comprised of real and synthetic data. The network trained on both real and synthetic datasets performed more accurate segmentation than the model trained only on real data. For instance, modified U-net trained on combined datasets performed segmentation with the Dice similarity coefficient of 92.6±2.2%, while the same model trained only on real samples achieved the level of 86.5±3.6%. Using a synthetic dataset allowed decreasing the accuracy spread and improving the generalization of the model. It is worth noting that the proposed algorithm allows reducing subjectivity, minimizing the labeling routine, increasing the number of samples, and improving the heterog

关键词： Echocardiography

来源：评论

学校读者我要写书评

暂无评论

Semi-supervised Segmentation and Quantization Algorithm for Infarct Size Measurement of Rat Heart Sections

Semi-supervised Segmentation and Quantization Algorithm for ...

引用

International Conference on Computer and Communications (ICCC)

作者： Xiaoke Zhu Yulong Cai Xiaopan Chen Qi Hu Lingyun Dong Caihong Yuan School of Computer and Information Engineering Henan University Kaifeng China Henan Engineering Research Center of Intelligent Technology and Application Henan University Henan Key Laboratory of Big Data Analysis and Processing Henan University Kaifeng China

Segmentation of rat heart section images and infarct area quantization are essential components of infarct size measurement in processing data from animal experiments on ischemic heart disease. Previously, the task of measuring the infarct size of rat heart sections was performed manually by experienced researchers, which was very costly and time-consuming. In this paper, we propose a two-stage semi-supervised approach to solve the task of ischemic heart infarct size calculation in images of rat section images. The approach uses a semi-supervised semantic segmentation model and a semi-automated quantization algorithm to separate the differently stained regions of rat heart sections. We demonstrate the results of the approach on a rat heart section dataset, and experiments on this dataset emphasize the effectiveness of our approach compared to other methods.

关键词： Heart Training Quantization (signal) Semantic segmentation Manuals Rats Size measurement

来源：评论

学校读者我要写书评

暂无评论

Syntax-enhanced pre-trained model 59

Syntax-enhanced pre-trained model

引用

Joint Conference of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language processing, ACL-IJCNLP 2021

作者： Xu, Zenan Guo, Daya Tang, Duyu Su, Qinliang Shou, Linjun Gong, Ming Zhong, Wanjun Quan, Xiaojun Jiang, Daxin Duan, Nan School of Computer Science and Engineering Sun Yat-Sen University Guangzhou China Microsoft Research Asia Beijing China Microsoft Search Technology Center Asia Beijing China Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou China Key Lab. of Machine Intelligence and Advanced Computing Ministry of Education China

ISBN: (纸本)9781954085527

We study the problem of leveraging the syntactic structure of text to enhance pre-trained models such as BERT and RoBERTa. Existing methods utilize syntax of text either in the pre-training stage or in the fine-tuning stage, so that they suffer from discrepancy between the two stages. Such a problem would lead to the necessity of having human-annotated syntactic information, which limits the application of existing methods to broader scenarios. To address this, we present a model that utilizes the syntax of text in both pre-training and fine-tuning stages. Our model is based on Transformer with a syntax-aware attention layer that considers the dependency tree of the text. We further introduce a new pre-training task of predicting the syntactic distance among tokens in the dependency tree. We evaluate the model on three downstream tasks, including relation classification, entity typing, and question answering. Results show that our model achieves state-of-the-art performance on six public benchmark datasets. We have two major findings. First, we demonstrate that infusing automatically produced syntax of text improves pre-trained models. Second, global syntactic distances among tokens bring larger performance gains compared to local head relations between contiguous tokens. © 2021 Association for Computational Linguistics

关键词： Syntactics

来源：评论

学校读者我要写书评

暂无评论

Point-Level and Set-Level Deep Representation Learning for Cross-Modal Person Re-identification

Point-Level and Set-Level Deep Representation Learning for C...

引用

International Conference on Computer and Communications (ICCC)

作者： Jihui Hu Pengfei Ye Danyang Li Lingyun Dong Xiaopan Chen Xiaoke Zhu Henan Key Laboratory of Big Data Analysis and Processing Henan University Kaifeng China School of Computer and Information Engineering Henan University Kaifeng China Henan Engineering Research Center of Intelligent Technology and Application Henan University

In practice, significant modality differences usually exist between visible and infrared images, which makes visible-infrared Person Re-Identification (VI-ReID) a challenging research task. Due to the existing influence of pose variation, background changes, and occlusion, there are often outlier samples within the set of images from the same person. These outlier samples can adversely affect the process of learning the cross-modal matching model. The existing VI-ReID methods mainly focus on learning cross-modal feature representation by using image-level discriminant constraints, i.e., the distance between the truly-matching cross-modal images should be smaller than that between wrong-matching cross-modal images. However, most of these methods ignore the adverse influence caused by outliers. To solve the above problems, we proposed a Point-level and Set-level Deep Representation Learning (PSDRL) approach for VI-ReID in this paper. By using the set-level constraint in the process of deep representation learning, the discrepancy between visible and infrared modalities can be decreased, and the adverse effect of outliers can be weakened. By employing the image-level constraint, the discriminability of the obtained deep feature representations can be improved. Extensive experiments are conducted on the publicly available cross-modal Person Re-Identification datasets, including SYSU-MM01 and RegDB. Experimental results demonstrate the effectiveness of the proposed approach.

关键词： Representation learning Training image color analysis Computational modeling Task analysis

来源：评论

学校读者我要写书评

暂无评论

Dual attention mechanism object tracking algorithm based on Fully-convolutional Siamese network

Dual attention mechanism object tracking algorithm based on ...

引用

2021 International Conference on Networking and Network Applications, NaNA 2021

作者： Ma, Sugang Zhang, Zixian Zhang, Lei Chen, Yanping Yang, Xiaobao Pu, Lei Hou, Zhiqiang Xi'an University of Posts and Telecommunications School of Computer Science and Technology Shaanxi Xi'an710121 China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Xi'an Key Laboratory of Big Data and Intelligent Computing Xi'an University of Posts and Telecommunications Shaanxi Xi'an710121 China School of Information and Navigation Air Force Engineering University Shaanxi Xi'an710077 China School of Information Engineering Chang'an University Shaanxi Xi'an710064 China

ISBN: (纸本)9781665441582

In an effort to the problem of insufficient tracking performance of the Fully-convolutional Siamese network (SiamFC) in complex scenarios, a dual attention mechanism object tracking algorithm based on the Fully-convolutional Siamese network is proposed to improve the generalization capability of the tracker by ameliorating the robustness of the template characteristics. Firstly, a global context attention module is appended after the backbone network of SiamFC to ameliorate the power of original feature extraction from two dimensions of spatial and channel. Then, a coordinate attention module is introduced to augment the capability of feature extraction in the channel dimension. Finally, the model of the proposed algorithm is trained on the Got-10k dataset. Five related algorithms are tested on the OTB2015 dataset, the results of experiments manifest that our algorithm outperforms the baseline trackers, the success and precision rate of the proposed algorithm are improved by 3.3% and 6.3%. The average tracking speed is 145FPS, which can demand the requirement of real-time tracking. © 2021 IEEE.

关键词： Feature extraction

来源：评论

学校读者我要写书评

暂无评论

Anomaly Detection by Leveraging Incomplete Anomalous Knowledge with Anomaly-Aware Bidirectional GANs

arXiv

引用

arXiv 2022年

作者： Tian, Bowen Su, Qinliang Yin, Jian School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou China School of Artificial Intelligence Sun Yat-sen University Guangdong China

The goal of anomaly detection is to identify anomalous samples from normal ones. In this paper, a small number of anomalies are assumed to be available at the training stage, but they are assumed to be collected only from several anomaly types, leaving the majority of anomaly types not represented in the collected anomaly dataset at all. To effectively leverage this kind of incomplete anomalous knowledge represented by the collected anomalies, we propose to learn a probability distribution that can not only model the normal samples, but also guarantee to assign low density values for the collected anomalies. To this end, an anomaly-aware generative adversarial network (GAN) is developed, which, in addition to modeling the normal samples as most GANs do, is able to explicitly avoid assigning probabilities for collected anomalous samples. Moreover, to facilitate the computation of anomaly detection criteria like reconstruction error, the proposed anomaly-aware GAN is designed to be bidirectional, attaching an encoder for the generator. Extensive experimental results demonstrate that our proposed method is able to effectively make use of the incomplete anomalous information, leading to significant performance gains compared to existing methods1 © 2022, CC BY.

关键词： Anomaly detection

来源：评论

学校读者我要写书评

暂无评论

Lightweight Transformer Network and Self-supervised Task for Kinship Verification

Lightweight Transformer Network and Self-supervised Task for...

引用

International Conference on Computer and Communications (ICCC)

作者： Xiaoke Zhu Yunwei Li Danyang Li Lingyun Dong Xiaopan Chen School of Computer and Information Engineering Henan University Kaifeng China Henan Engineering Research Center of Intelligent Technology and Application Henan University Henan Key Laboratory of Big Data Analysis and Processing Henan University Kaifeng China

Kinship verification is one of the interesting and critical problems in computer vision research, with significant progress in the past decades. Meanwhile, Vision Transformer (VIT) has recently achieved impressive success in many domains, including object detection, image recognition, and semantic segmentation, among others. Most of the previous work on kinship verification are based on convolutional or recurrent neural networks. Compared with the local processing of images like convolutions, transformers can effectively understand and process images globally. However, due to overuse, there are many Transformer layers of fully connected layers, and VIT speed is still an issue. Therefore, in this paper, inspired by the recent success of Transformer models in vision tasks, we propose a Transformer-based kinship verification for training and optimizing kinship verification models. We first train the basic vision transformer (VIT-B) with 12 transformer layers, then we reduce the transformer layers to 6 layers, namely VIT-S (Small Vit) and 4 layers, namely VIT-T (Tiny Vit), to make a tradeoff between optimization accuracy and efficiency. As the first attempt to apply Transformer to the kinship verification task, it provides a feasible strategy for kinship research topics and verifies the effectiveness of the method in terms of the accuracy of the experimental results.

关键词： Training Knowledge engineering Recurrent neural networks Semantic segmentation Self-supervised learning Object detection Transformers

来源：评论

学校读者我要写书评

暂无评论

A High Parallel Accelerator Design Based on the Sparsity of Convolutional Neural Network

SSRN

引用

SSRN 2022年

作者： Pan, Xiao Ying Mu, Yuanzhen Jia, Ningxin Gao, Xuan Rong Feng, Tong School of Computer Science and Technology Xi’an University of Posts and Telecommunications Shannxi Xi’an710121 China Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Xi’an University of Posts and Telecommunications Shannxi Xi’an710121 China

Designing a domain-specific accelerator is an effective way to improve the performance of inference network on resource-constrained embedded devices. However, there are a large number of invalid memory accesses and calculations in the convolutional neural network. Based on this, this paper proposes an accelerator that adopts a flexible neural network structure that adjusts the processing flow of the network, reduces the size of the feature map sent into the activation function, and greatly reduces the inherent invalid calculation and memory access of the network. The accelerator uses loop expansion and loop reconstruction to solve the read-write problem in convolution operation, realize instruction-level parallelism and data-level parallelism. We determine the design parameters of parallel computing by modelling the constrained single-objective optimization problem to find the optimal cyclic expansion factor. Evaluate the performance, resource utilization, and energy consumption of accelerators in different neural networks. Compared with the general processor CPU and GPU, the running time of the accelerator is shortened by 6.89s and 1.16s, and the power consumption is only 7.47% of CPU and 3% of GPU. Compared with the state-of-the-art accelerator, the performance is increased by 2.8x and the power consumption is reduced by approximately 6W. © 2022, The Authors. All rights reserved.

关键词： Field programmable gate arrays (FPGA)

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：