DV-Hop localization algorithm contains a straightforward structure and is widely employed in wireless sensor network node localization. The DV-Hop localization algorithm has a fundamental error within the average hop distance calculation. Thus this paper proposes a DV-Hop localization algorithm using classifying average hop distance(CADV-Hop algorithm). We provide a CADV-Hop algorithm to calculate the typical hop distance by categorizing utterly different hop counts. The simulation results show that the improved algorithm will improve the localization accuracy compared to the DV-Hop algorithm.
{"title":"A DV-Hop Localization Algorithm Using Classifying Average Hop Distance in Wireless Sensor Networks","authors":"Di Yang, Xuanzhi Zhao, Wenpeng Zhang","doi":"10.1145/3569966.3570030","DOIUrl":"https://doi.org/10.1145/3569966.3570030","url":null,"abstract":"DV-Hop localization algorithm contains a straightforward structure and is widely employed in wireless sensor network node localization. The DV-Hop localization algorithm has a fundamental error within the average hop distance calculation. Thus this paper proposes a DV-Hop localization algorithm using classifying average hop distance(CADV-Hop algorithm). We provide a CADV-Hop algorithm to calculate the typical hop distance by categorizing utterly different hop counts. The simulation results show that the improved algorithm will improve the localization accuracy compared to the DV-Hop algorithm.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128303305","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In order to avoid the situation where the weight assignment value is zero in the data envelopment analysis, the weights of its input and output indexes should be restricted. In this paper, the objective weight determined by the anti-entropy method is combined with the subjective weight determined by the Delphi method, and then the combined weight is obtained according to the principle of minimum variance, which is added to the data envelopment analysis as a constraint condition to construct the super efficiency DEA model with anti-entropy-Delphi combined weights constrains cone. The new evaluation model with the constraint cone can not only reflect the objective impact of data on the index weight, but also integrate the subjective consciousness of experts, and achieve a complete ranking of the evaluation results. Finally, an empirical analysis of the innovation efficiency of basic research in Beijing from 2011 to 2020 is made based on panel data. The result shows that the new model has obvious advantages compared with the super-efficiency DEA model.
{"title":"Super Efficiency DEA Evaluation Method with Anti-Entropy-Delphi Combined Weights Constraints Cone","authors":"Na Xu","doi":"10.1145/3569966.3571192","DOIUrl":"https://doi.org/10.1145/3569966.3571192","url":null,"abstract":"In order to avoid the situation where the weight assignment value is zero in the data envelopment analysis, the weights of its input and output indexes should be restricted. In this paper, the objective weight determined by the anti-entropy method is combined with the subjective weight determined by the Delphi method, and then the combined weight is obtained according to the principle of minimum variance, which is added to the data envelopment analysis as a constraint condition to construct the super efficiency DEA model with anti-entropy-Delphi combined weights constrains cone. The new evaluation model with the constraint cone can not only reflect the objective impact of data on the index weight, but also integrate the subjective consciousness of experts, and achieve a complete ranking of the evaluation results. Finally, an empirical analysis of the innovation efficiency of basic research in Beijing from 2011 to 2020 is made based on panel data. The result shows that the new model has obvious advantages compared with the super-efficiency DEA model.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128363078","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
load forecasting is an important research direction, which has always been the concern of academia and industry. Accurate prediction results can provide effective decisions for resource allocation of the system. However, the change of application load is very complex. How to accurately predict the change trend of load is a challenging task. Traditional prediction algorithms such as Arima algorithm based on statistical theory and neural network algorithm predict the target load only through the historical sequence of a single load index, ignoring the interaction between different load indexes. Therefore, this paper proposes a load prediction model based on long-term and short-term memory network and attention mechanism lstmda. The model successively uses convolutional neural network and channel attention mechanism to extract the local dependence characteristics between loads. The bidirectional LSTM network with attention mechanism is used to predict the load, and the data at different times are given different degrees of importance. The model proposed in this paper achieves better performance than existing prediction algorithms on real load data sets.
{"title":"Load prediction model based on LSTM and attention mechanism","authors":"Xuan Zhou, Xing Wu","doi":"10.1145/3569966.3570095","DOIUrl":"https://doi.org/10.1145/3569966.3570095","url":null,"abstract":"load forecasting is an important research direction, which has always been the concern of academia and industry. Accurate prediction results can provide effective decisions for resource allocation of the system. However, the change of application load is very complex. How to accurately predict the change trend of load is a challenging task. Traditional prediction algorithms such as Arima algorithm based on statistical theory and neural network algorithm predict the target load only through the historical sequence of a single load index, ignoring the interaction between different load indexes. Therefore, this paper proposes a load prediction model based on long-term and short-term memory network and attention mechanism lstmda. The model successively uses convolutional neural network and channel attention mechanism to extract the local dependence characteristics between loads. The bidirectional LSTM network with attention mechanism is used to predict the load, and the data at different times are given different degrees of importance. The model proposed in this paper achieves better performance than existing prediction algorithms on real load data sets.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127331907","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper uses a web crawler program to collect 2800 review texts of Duxiufeng Wangcheng Scenic Area on Ctrip.com, and after processing, 2725 review texts are retained as research samples, combined with text analysis methods, using ROSTCM6 for sentiment analysis, word cloud graph analysis and semantic network analysis to analyze user review sentiment. It is found that users have mostly positive attitudes towards Duxiufeng Wangcheng scenic spot reviews, but there are large differences between different reviewers' reviews; users are more concerned about scenic spots, guided tours, entrance fees and entertainment program costs; the scenic spot is an experience-based comprehensive scenic spot. And based on the above results, we propose countermeasures for optimizing the development of Guilin scenic spot tourism.
{"title":"Research on Tourists' Emotional Expression Based on Web Text Analysis","authors":"Ling Xiao, Fuxi Liang, Kaiyong Cheng, Huiru Xu","doi":"10.1145/3569966.3570082","DOIUrl":"https://doi.org/10.1145/3569966.3570082","url":null,"abstract":"This paper uses a web crawler program to collect 2800 review texts of Duxiufeng Wangcheng Scenic Area on Ctrip.com, and after processing, 2725 review texts are retained as research samples, combined with text analysis methods, using ROSTCM6 for sentiment analysis, word cloud graph analysis and semantic network analysis to analyze user review sentiment. It is found that users have mostly positive attitudes towards Duxiufeng Wangcheng scenic spot reviews, but there are large differences between different reviewers' reviews; users are more concerned about scenic spots, guided tours, entrance fees and entertainment program costs; the scenic spot is an experience-based comprehensive scenic spot. And based on the above results, we propose countermeasures for optimizing the development of Guilin scenic spot tourism.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129087760","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Sentence representation learning can transform sentences into fixed format vectors, and provides foundation for downstream tasks such as information retrieval, semantic similarity analysis, etc. With the popularity of contrastive learning, sentence representation learning has also been further developed. At the same time, contrastive learning method based on momentum has achieved great success in computer vision. It solves the coupling between negative samples and batch size. But its expected performance is not observed in natural language processing tasks because the combination of data augmentation strategies is weak, and it only utilizes the samples in the momentum queue as negatives while ignoring those generated in current batch. In this paper, we propose eMoCo: enhanced Momentum Contrast to solve the above issues. We formulate a set of data augmentation strategies for text, and present a novel Dual-Negative loss to make full use of all negative samples. Extensive experiments on STS (Semantic Text Similarity) datasets show that our method outperforms the current state-of-the-art models, indicating its advantages in sentence representation learning.
{"title":"eMoCo: Sentence Representation Learning With Enhanced Momentum Contrast","authors":"Shibo Qi, Rize Jin, Joon-Young Paik","doi":"10.1145/3569966.3570013","DOIUrl":"https://doi.org/10.1145/3569966.3570013","url":null,"abstract":"Sentence representation learning can transform sentences into fixed format vectors, and provides foundation for downstream tasks such as information retrieval, semantic similarity analysis, etc. With the popularity of contrastive learning, sentence representation learning has also been further developed. At the same time, contrastive learning method based on momentum has achieved great success in computer vision. It solves the coupling between negative samples and batch size. But its expected performance is not observed in natural language processing tasks because the combination of data augmentation strategies is weak, and it only utilizes the samples in the momentum queue as negatives while ignoring those generated in current batch. In this paper, we propose eMoCo: enhanced Momentum Contrast to solve the above issues. We formulate a set of data augmentation strategies for text, and present a novel Dual-Negative loss to make full use of all negative samples. Extensive experiments on STS (Semantic Text Similarity) datasets show that our method outperforms the current state-of-the-art models, indicating its advantages in sentence representation learning.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130204512","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Abstract—In order to reveal the impact of Covid-19 on special work of chemical production enterprises in China, including Chinese funded, Sino foreign joint venture, foreign funded and foreign funded joint venture enterprises. The mutual influences between the two were studied by collecting safety commitment and Covid-19 logs in Zhejiang Province, and correlation method and event study were adopted. Although the level of Covid-19 was lowered to level 3 on March 23 by Zhejiang Province, it would take 46 days, 31 days, 61 days and 46 days for the four types to return to the normal special work level respectively. Therefore, for similar major public health events, chemical production enterprises should timely avoid their impact on special operation activities related to normal inspections and maintenances.
{"title":"Application of analysis on the impact of major public health events on special work of chemical enterprises","authors":"Yin-gang Wu","doi":"10.1145/3569966.3571185","DOIUrl":"https://doi.org/10.1145/3569966.3571185","url":null,"abstract":"Abstract—In order to reveal the impact of Covid-19 on special work of chemical production enterprises in China, including Chinese funded, Sino foreign joint venture, foreign funded and foreign funded joint venture enterprises. The mutual influences between the two were studied by collecting safety commitment and Covid-19 logs in Zhejiang Province, and correlation method and event study were adopted. Although the level of Covid-19 was lowered to level 3 on March 23 by Zhejiang Province, it would take 46 days, 31 days, 61 days and 46 days for the four types to return to the normal special work level respectively. Therefore, for similar major public health events, chemical production enterprises should timely avoid their impact on special operation activities related to normal inspections and maintenances.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"102 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130542857","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Xinyi Xie, Haibin Zheng, Hu Li, Ling Pang, Jinyin Chen
To further protect citizens' privacy and national data security, graph federation learning has been widely used and rapidly developed. However, with the deployment and landing of graph federation learning tasks, the security issues involved are gradually exposed. To deeply study the application security issues of graph federation learning, this paper proposes an attack method and privacy protection defense method for graph data in the framework of vertical federation learning. The research revolves around the attack method. First, noise is randomly generated, combined with the attacker's embedding features, and fed into the server model, and the calculated results are compared with the real values to obtain the loss values. Then the attacker's attack model is updated to generate a new inference of the attacked embedding. The experiments conducted on two real-world datasets both obtained MSE metrics below 1, which fully illustrates the effectiveness of the attack method. Further research is conducted around the defense method, which uses a computed differential noise added to the uploaded embedding information to achieve the defense against privacy theft. In the experiments, the related attack metrics are significantly reduced with almost no impact on the main task performance.
{"title":"Attack and Defense Methods for Graph Vertical Federation Learning","authors":"Xinyi Xie, Haibin Zheng, Hu Li, Ling Pang, Jinyin Chen","doi":"10.1145/3569966.3570022","DOIUrl":"https://doi.org/10.1145/3569966.3570022","url":null,"abstract":"To further protect citizens' privacy and national data security, graph federation learning has been widely used and rapidly developed. However, with the deployment and landing of graph federation learning tasks, the security issues involved are gradually exposed. To deeply study the application security issues of graph federation learning, this paper proposes an attack method and privacy protection defense method for graph data in the framework of vertical federation learning. The research revolves around the attack method. First, noise is randomly generated, combined with the attacker's embedding features, and fed into the server model, and the calculated results are compared with the real values to obtain the loss values. Then the attacker's attack model is updated to generate a new inference of the attacked embedding. The experiments conducted on two real-world datasets both obtained MSE metrics below 1, which fully illustrates the effectiveness of the attack method. Further research is conducted around the defense method, which uses a computed differential noise added to the uploaded embedding information to achieve the defense against privacy theft. In the experiments, the related attack metrics are significantly reduced with almost no impact on the main task performance.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123292655","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
For the current problems of low detection accuracy and slow detection speed of PCB board defect detection, this paper proposes a method of PCB defect detection by YOLOv5 algorithm with Transformer module added. The algorithm is using Transformer encoder block to replace some convolution blocks and bottleneck blocks in YOLOv5. it uses the self-attention mechanism to tap the feature representation potential and solve the problem of low resolution of the feature map at the end of the network. The experimental results show that the improved algorithm can better identify the defects of PCB boards, the detection accuracy mAP reaches 97.8%, and the average detection time is improved from 194.2ms to 183.5ms. it is suitable for the actual production and inspection process.
{"title":"Method of PCB defect detection with yolov5 algorithm by adding transformer module","authors":"Yuqing Li, Zuguo Chen","doi":"10.1145/3569966.3570054","DOIUrl":"https://doi.org/10.1145/3569966.3570054","url":null,"abstract":"For the current problems of low detection accuracy and slow detection speed of PCB board defect detection, this paper proposes a method of PCB defect detection by YOLOv5 algorithm with Transformer module added. The algorithm is using Transformer encoder block to replace some convolution blocks and bottleneck blocks in YOLOv5. it uses the self-attention mechanism to tap the feature representation potential and solve the problem of low resolution of the feature map at the end of the network. The experimental results show that the improved algorithm can better identify the defects of PCB boards, the detection accuracy mAP reaches 97.8%, and the average detection time is improved from 194.2ms to 183.5ms. it is suitable for the actual production and inspection process.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"64 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122286163","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
According to previous research, sunspots and weather conditions have both direct and latent economic impacts, such as human financial activities. The goal of this project was to see if machine learning and deep neural network methods could reveal a link between natural phenomena, specifically sunspots, weather, and agricultural stock price trends. I suggested that some of these natural events could be related to the price trends of individual equities. Using machine learning and deep neural network methods, I analysed at both the general Dow Jones index level and the particular agriculture stock level. Outperforming other models, the LSTM (Long-Short-Term Memory) model produced an MSE (Mean Squared Error) error of 9.91 between the sunspot number and various agricultural price patterns, which was far lower than my hypothesis. The outcome shifts from standard algorithm trading to a completely new aspect, with (space) meteorological factors playing critical roles for the first time. The implications of these results extended far beyond commercial advantages. The findings provided unique proof that not only our commercial world is impacted by space weather, the impact of which can also be digitally recorded and anticipated. This preliminary but effective study established a computer link between space weather and human business behavior, sparking one's vivid imagination of the forces at work.
根据之前的研究，太阳黑子和天气状况对经济有直接和潜在的影响，比如人类的金融活动。这个项目的目标是看看机器学习和深度神经网络方法是否可以揭示自然现象之间的联系，特别是太阳黑子、天气和农业股票价格趋势。我认为，其中一些自然事件可能与个别股票的价格趋势有关。使用机器学习和深度神经网络方法，我分析了道琼斯指数水平和特定农业股票水平。LSTM (long - short - short Memory，长短期记忆)模型在太阳黑子数量与各种农产品价格模式之间产生的MSE(均方误差)误差为9.91，远低于我的假设，优于其他模型。结果从标准的算法交易转向了一个全新的方面，(空间)气象因素首次发挥了关键作用。这些结果的含义远远超出了商业优势。这些发现提供了独特的证据，证明不仅我们的商业世界受到太空天气的影响，其影响也可以通过数字记录和预测。这项初步但有效的研究在太空天气和人类商业行为之间建立了计算机联系，激发了人们对工作力量的生动想象。
{"title":"Trends in Sunspots & Agriculture Stock Prices - Finding Correlations and Predicting Trends Using Machine Learning and Deep Neural Networks","authors":"Kwan Yeung Liu","doi":"10.1145/3569966.3570098","DOIUrl":"https://doi.org/10.1145/3569966.3570098","url":null,"abstract":"According to previous research, sunspots and weather conditions have both direct and latent economic impacts, such as human financial activities. The goal of this project was to see if machine learning and deep neural network methods could reveal a link between natural phenomena, specifically sunspots, weather, and agricultural stock price trends. I suggested that some of these natural events could be related to the price trends of individual equities. Using machine learning and deep neural network methods, I analysed at both the general Dow Jones index level and the particular agriculture stock level. Outperforming other models, the LSTM (Long-Short-Term Memory) model produced an MSE (Mean Squared Error) error of 9.91 between the sunspot number and various agricultural price patterns, which was far lower than my hypothesis. The outcome shifts from standard algorithm trading to a completely new aspect, with (space) meteorological factors playing critical roles for the first time. The implications of these results extended far beyond commercial advantages. The findings provided unique proof that not only our commercial world is impacted by space weather, the impact of which can also be digitally recorded and anticipated. This preliminary but effective study established a computer link between space weather and human business behavior, sparking one's vivid imagination of the forces at work.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"28 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116612465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Human pose estimation methods have developed rapidly in recent years and many high precision models have emerged. However, the computational costs of these methods are often very huge, especially for transformer-based models. In this work, we propose ConvPose, an efficient human pose estimation model based on convolutional neural network architecture. ConvPose uses an efficient single branch structure, using the ConvNeXt Block as a baseline and incorporating the Coordinate Attention module. This composition not only provides better feature extraction capabilities, but also can efficiently obtain the global dependency relationships between human keypoints and scenes. The effective combination of the large convolution kernel and the attention module gives our model the ability to focus more on detailed features when oriented to complex scenes. In addition, the number of parameters and GFLOPs of our model are at a lighter level compared to current high- performance models, which offers more possibilities for deployment of the model in low-end devices. Experiments show that our model achieves 73.6AP on the MS-COCO dataset with only 6.3M parameters, which is a very competitive result.
{"title":"ConvPose: An efficient human pose estimation method based on ConvNeXt","authors":"Ke Lin, S. Zhang, Zhisong Qin","doi":"10.1145/3569966.3569989","DOIUrl":"https://doi.org/10.1145/3569966.3569989","url":null,"abstract":"Human pose estimation methods have developed rapidly in recent years and many high precision models have emerged. However, the computational costs of these methods are often very huge, especially for transformer-based models. In this work, we propose ConvPose, an efficient human pose estimation model based on convolutional neural network architecture. ConvPose uses an efficient single branch structure, using the ConvNeXt Block as a baseline and incorporating the Coordinate Attention module. This composition not only provides better feature extraction capabilities, but also can efficiently obtain the global dependency relationships between human keypoints and scenes. The effective combination of the large convolution kernel and the attention module gives our model the ability to focus more on detailed features when oriented to complex scenes. In addition, the number of parameters and GFLOPs of our model are at a lighter level compared to current high- performance models, which offers more possibilities for deployment of the model in low-end devices. Experiments show that our model achieves 73.6AP on the MS-COCO dataset with only 6.3M parameters, which is a very competitive result.","PeriodicalId":145580,"journal":{"name":"Proceedings of the 5th International Conference on Computer Science and Software Engineering","volume":"3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125298952","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：