检索结果-内蒙古大学图书馆

HyRSM++: Hybrid relation guided temporal set matching for few-shot action recognition

Pattern Recognition 2024年 147卷

作者： Wang, Xiang Zhang, Shiwei Qing, Zhiwu Zuo, Zhengrong Gao, Changxin Jin, Rong Sang, Nong Key Laboratory of Ministry of Education for Image Processing and Intelligent Control School of Artificial Intelligence and Automation Huazhong University of Science and Technology China Alibaba Group China Meta AI United States

Few-shot action recognition is a challenging but practical problem aiming to learn a model that can be easily adapted to identify new action categories with only a few labeled samples. However, existing attempts still suffer from two drawbacks: (i) learning individual features without considering the entire task may result in limited representation capability, and (ii) existing alignment strategies are sensitive to noises and misaligned instances. To handle the two limitations, we propose a novel Hybrid Relation guided temporal Set Matching (HyRSM++) approach for few-shot action recognition. The core idea of HyRSM++ is to integrate all videos within the task to learn discriminative representations and involve a robust matching technique. To be specific, HyRSM++ consists of two key components, a hybrid relation module and a temporal set matching metric. Given the basic representations from the feature extractor, the hybrid relation module is introduced to fully exploit associated relations within and cross videos in an episodic task and thus can learn task-specific embeddings. Subsequently, in the temporal set matching metric, we carry out the distance measure between query and support videos from a set matching perspective and design a bidirectional Mean Hausdorff Metric to improve the resilience to misaligned instances. Furthermore, we extend the proposed HyRSM++ to deal with the more challenging semi-supervised few-shot action recognition and unsupervised few-shot action recognition tasks. Experimental results on multiple benchmarks demonstrate that our method consistently outperforms existing methods and achieves state-of-the-art performance under various few-shot settings. The source code is available at https://***/alibaba-mmai-research/HyRSMPlusPlus. © 2023 Elsevier Ltd

关键词： Benchmarking

来源：评论

学校读者我要写书评

暂无评论

A Recipe for Scaling up Text-to-Video Generation with Text-free Videos

A Recipe for Scaling up Text-to-Video Generation with Text-f...

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Xiang Wang Shiwei Zhang Hangjie Yuan Zhiwu Qing Biao Gong Yingya Zhang Yujun Shen Changxin Gao Nong Sang Key Laboratory of Image Processing and Intelligent Control School of Artificial Intelligence and Automation Huazhong University of Science and Technology Alibaba Group Zhejiang University Ant Group

ISBN: (数字)9798350353006

ISBN: (纸本)9798350353013

Diffusion-based text-to-video generation has witnessed impressive progress in the past year yet still falls behind text-to-image generation. One of the key reasons is the limited scale of publicly available data (e.g., 10M video-text pairs in WebVid10m vs. 5B image-text pairs in LAION), considering the high cost of video captioning. Instead, it could be far easier to collect unlabeled clips from video platforms like YouTube. Motivated by this, we come up with a novel text-to-video generation framework, termed TF-T2V, which can directly learn with text-free videos. The rationale behind is to separate the process of text decoding from that of temporal modeling. To this end, we employ a content branch and a motion branch, which are jointly optimized with weights shared. Following such a pipeline, we study the effect of doubling the scale of training set (i.e., video-only WebVid10M) with some randomly collected text-free videos and are encouraged to observe the performance improvement (FID from 9.67 to 8.19 and FVD from 484 to 441), demonstrating the scalability of our approach. We also find that our model could enjoy sustainable performance gain (FID from 8.19 to 7.64 and FVD from 441 to 366) after reintroducing some text labels for training. Finally, we validate the effectiveness and generalizability of our ideology on both native text-to-video generation and compositional video synthesis paradigms. Code and models will be publicly available at here.

关键词： Training Video on demand Scalability Pipelines Text to image Performance gain Market research

来源：评论

学校读者我要写书评

暂无评论

Selection of Optimal Focusing Evaluation Function Based on Sparse Microscopic images

Selection of Optimal Focusing Evaluation Function Based on S...

引用

2022 International Conference on Electrical, Electronics and Information Engineering, EEIE 2022

作者： Lv, Meini Gan, Hui Liu, Xin Chen, Jia Yang, Qiuhui Wuzhou University Wuzhou China Guangxi Colleges and Universities Key Laboratory of Image Processing and Intelligent Information System Wuzhou University Wuzhou China

The auto focusing process is affected by the amount of image content, resulting in the focus curve not meeting the characteristics of the ideal focus curve. In this paper, aiming at the problem that the traditional autofocus cannot find the focal plane successfully in the case of microscopic sparse content, the qualitative analysis and quantitative analysis of 12 traditional focus evaluation functions are carried out. The experimental results show that the F Tenengrad function is suitable for the optimal focus evaluation function of sparse microscopic images. © Published under licence by IOP Publishing Ltd.

关键词： Focusing

来源：评论

学校读者我要写书评

暂无评论

Ensemble prediction modeling of flotation recovery based on machine learning

引用

International Journal of Mining Science and Technology 2024年第12期34卷 1727-1740页

作者： Guichun He Mengfei Liu Hongyu Zhao Kaiqi Huang Jiangxi Provincial Key Laboratory of Low-Carbon Processing and Utilization of Strategic Metal Mineral Resources Ganzhou 341000China School of Resources and Environmental Engineering Jiangxi University of Science and TechnologyGanzhou 341000China Hefei GoldStart Intelligent Control Technical Co. Ltd.Hefei 230088China School of Mechanical and Electrical Engineering Jiangxi University of Science and TechnologyGanzhou 341000China

With the rise of artificial intelligence(AI)in mineral processing,predicting the flotation indexes has attracted significant research ***,current prediction models suffer from low accuracy and high prediction ***,this paper utilizes a two-step ***,the outliers are pro-cessed using the box chart method and filtering ***,the decision tree(DT),support vector regression(SVR),random forest(RF),and the bagging,boosting,and stacking integration algorithms are employed to construct a flotation recovery prediction *** experiments compared the prediction accuracy of six modeling methods on flotation recovery and delved into the impact of diverse base model combinations on the stacking model’s prediction *** addition,field data have veri-fied the model’s *** study demonstrates that the stacking ensemble approaches,which uses ten variables to predict flotation recovery,yields a more favorable prediction effect than the bagging ensemble approach and single models,achieving MAE,RMSE,R2,and MRE scores of 0.929,1.370,0.843,and 1.229%,*** hit rates,within an error range of±2%and±4%,are 82.4%and 94.6%.Consequently,the prediction effect is relatively precise and offers significant value in the context of actual production.

关键词： Machine learning Stacking Bagging Flotation recovery rate Filtering algorithm

来源：评论

学校读者我要写书评

暂无评论

Multi-Branch Spatial-Temporal Attention Graph Convolution Network for Skeleton-based Action Recognition 41

Multi-Branch Spatial-Temporal Attention Graph Convolution Ne...

引用

第41届中国控制会议

作者： Daoshuai Wang Dewei Li Yaonan Guan Gang Wang Haibin Shao Department of Automation Shanghai Jiao Tong University Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai Engineering Research Center of Intelligent Control and Management

ISBN: (数字)9789887581536

ISBN: (纸本)9781665482561

Skeleton-based human action reco gnition has attracted considerable research interest due to its robustness to dynamic environments and complex *** based on graph neural networks have achieved great success in this ***,most of these graph neural network based methods adopt a single-branch or two-stream structure with limited input *** to the separation of space and time,many methods cannot pay attention to the critical information of human skeleton sequence,and multi-scale spatial-temporal dependent attention modeling becomes the key to *** paper proposes a novel multi-branch Spatio-temporal attention graph convolutional neural network(MB-STAGCN) to recognize human actions from skeleton *** proposed model adopts a multi-branch structure,fuses three-stream features,and then inputs them into the backbone network to extract features from multiple perspectives to a greater ***,we design a new temporal convolutional block for multi-scale extraction of information in the backbone *** also propose an attention block named Spatio-temporal Concat Attention(STCatAtt) to capture critical *** on benchmark datasets show remarkable performance for human action reco gnition,demonstrating the effectiveness of our method.

关键词： Skeleton-based human action recognition MB-STAGCN STCatAtt

来源：评论

学校读者我要写书评

暂无评论

A Low-complexity Decision Variables Classification Brainstorm Optimization Algorithm

A Low-complexity Decision Variables Classification Brainstor...

引用

2024 International Conference on Applied Mathematics, Modelling and Statistics Application, AMMSA 2024

作者： Li, Qiyue Qiu, Lipeng Zhou, Aoran Wu, Yali Wang, Ziren Department of Information and Control Engineering Xi'An University of Technolog Xi'an710048 China Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing Xi'an710048 China State Key Laboratory of Metal Extrusion and Forging Equipment Technology China National Heavy Machinery Research Institute Co. Ltd. Xi'an710016 China

In the complex optimization scenarios of real life, we often need to not only weigh the conflicts between multi-objectives but also face the challenges brought by large-scale decision variables. When the scale of decision variables increases sharply, the search space will expand exponentially, which makes it difficult for the algorithm to traverse the whole search space under limited resources, so that the number of local optimal solutions increases sharply. The so-called 'dimension disaster' appears. To effectively deal with this problem, this paper proposes a large-scale multi-objective brainstorm optimization algorithm (LMaOBSO) based on low complexity decision variable classification method and improved penalty boundary intersection strategy. The algorithm uses a low-complexity tree sorting variable classification method to divide the decision variables into converges and diversity variables quickly. At the same time, the convergent variables are optimized by the improved penalty boundary crossover strategy to solve the problems existing in large-scale optimization problems. © Published under licence by IOP Publishing Ltd.

关键词： Disasters

来源：评论

学校读者我要写书评

暂无评论

Learning Knowledge Evolution with Time Duration from Finer-Time-Granularity Temporal Knowledge Graph

Learning Knowledge Evolution with Time Duration from Finer-T...

引用

2023 China Automation Congress, CAC 2023

作者： Du, Shaochong Chen, Kang Yang, Shijie Huo, Hong Fang, Tao Shanghai Jiao Tong University Department of Automation Shanghai China Ministry of Education Key Laboratory of System Control and Information Processing Shanghai China Shanghai Engineering Research Centre of Intelligent Control and Management Shanghai China

ISBN: (纸本)9798350303759

How to represent the temporal information corresponding to each fact in temporal knowledge graphs (TKGs) effectively is always challenging. Most existing representation learning methods usually map the timelines of knowledges into multiple points and ignore the underlying continuous time property and periodic characteristics behind those dense timestamps in finer-time granularity temporal knowledge graphs. A novel Knowledge Evolution with Time Duration (KETD) model is proposed for knowledge representation of temporal knowledge graphs with finer-time granularities. It represents the entity embedding as a time-varying nonlinear function to find the specific continuous time property from dense timestamps. To the best of our knowledge, it is the first attempt to learn embeddings of time durations and combines them into embeddings of entities and relations to predict potential periodic facts. It also uses the temporal point process to capture the impact of historical facts on the current fact. The experiments on self-built finer-time granularity TKG of Satellite-to-Earth Communication (STEC) and two public datasets have demonstrated the superiority of KETD compared to some baseline approaches. © 2023 IEEE.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

An Integrated Scheduling Algorithm of Satellite Observation and Data Download Based on Multi-Agent Deep Reinforcement Learning

An Integrated Scheduling Algorithm of Satellite Observation ...

引用

2023 China Automation Congress, CAC 2023

作者： Chen, Kang Du, Shaochong Huo, Hong Fang, Tao Shanghai Jiao Tong University Department of Automation Shanghai China Key Laboratory of System Control and Information Processing Ministry of Education Shanghai China Shanghai Engineering Research Centre of Intelligent Control and Management Shanghai China

ISBN: (纸本)9798350303759

The earth observation satellite scheduling has always been critical for the maximum use of limited satellite resources, which basically includes scheduling ground target observation and observation data downloading. Due to the complexity of the mathematical model and its NP-hardness, current research usually splits it into two problems to simplify the model. In this paper, we propose a model of the integrated scheduling problem of satellite observation and data downloading (ISP-SODD) based on the decentralized partially observable Markov decision process (Dec-POMDP), treating each satellite as an agent with autonomous decision-making capability. A multi-agent deep reinforcement learning-based (MADRL-based) algorithm is proposed to solve it. A series of simulation experiments of different scales validate the effectiveness of the method and its advantages compared with the traditional heuristic algorithms. To the best of our knowledge, this study is the first attempt to apply multi-agent reinforcement learning to a multi-satellite scheduling scenario with downloading considerations. © 2023 IEEE.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Small Object Detection via a Dense Connection and Feature Enhancement Network 9

Small Object Detection via a Dense Connection and Feature En...

引用

9th International Conference on Computer and Communications, ICCC 2023

作者： Qiang, Baohua Chen, Lirui Guo, Shuiping Liu, Lingzhi Zhang, Shihao Guilin University of Electronic Technology Guangxi Key Laboratory of Image and Graphic Intelligent Processing Guilin541004 China The 7th Research Institute of China Electrics Technology Group Corporation Guangzhou China

ISBN: (纸本)9798350317251

When detecting small objects in complex environment, the features of small objects will become blurred or even lost as the number of network layers increases. To address this problem, we constructed a Dense Connection and Feature Extraction Network, termed DCFEN. First, we designed a dense-connected multi-scale feature enhancement framework, which can effectively extract and fuse multi-scale features. Second, we constructed a dense-connected subnet using dense connections, which enhances the propagation of features and the utilization of shallow feature information, improving the detection performance of small objects. Finally, extensive experimental results demonstrated the detection precision and superiority of our method. © 2023 IEEE.

关键词： Network layers

来源：评论

学校读者我要写书评

暂无评论

Reward Shaping-based Double Deep Q-networks for Unmanned Surface Vessel Navigation and Obstacle Avoidance 48

Reward Shaping-based Double Deep Q-networks for Unmanned Sur...

引用

48th Annual Conference of the IEEE Industrial Electronics Society, IECON 2022

作者： Gan, Zihan Zheng, Jinghong Jiang, Zhenyu Lu, Renzhi Huazhong University of Science and Technology Key Laboratory of Image Processing and Intelligent Control School of Artificial Intelligence and Automation Wuhan430074 China Wuhan University of Science and Technology Hubei Key Laboratory of Mechanical Transmission and Manufacturing Engineering Wuhan430081 China

ISBN: (数字)9781665480253

ISBN: (纸本)9781665480253

In this paper, a method for navigation and obstacle avoidance of unmanned surface vessel (USV) based on reinforcement learning and reward shaping is proposed. This approach uses double deep Q networks (DDQN) to make decisions based on the continuous states observed from sensors in USV. In addition, a new reward function is designed based on prior knowledge to accelerate the convergence of the algorithm and improve the performance. For training the neural networks, a simulation platform is developed, in which a 3 degree of freedom mathematical model describes USV dynamic system and two-dimension actions are required to control USV. Simulation results on the platform demonstrate the DDQN hoists USV's capabilities of navigation and obstacle avoidance, and reward shaping technique improves the speed of convergence. © 2022 IEEE.

关键词： Reinforcement learning

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：