Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185515
T. Hamada, S. Itai
This article proposes a robot recreation network system to promote robot recreation. The network system consists of a base center, which has many robots, a robot engineer, and branches such as nursing homes. The members of the base center bring robots to the branch and carry out robot recreation there.
{"title":"Proposal of Robot Recreation Network System","authors":"T. Hamada, S. Itai","doi":"10.1109/UV56588.2022.10185515","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185515","url":null,"abstract":"This article proposes a robot recreation network system to promote robot recreation. The network system consists of a base center, which has many robots, a robot engineer, and branches such as nursing homes. The members of the base center bring robots to the branch and carry out robot recreation there.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"76 3 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116309904","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185481
Zihan Xie, Dongquan Li
With the development of the times, information technology is used to build smart communities, which is considered to be the way to improve community governance ability and meet the growing needs of residents for a better life. However, whether the technological revolution can achieve the people-oriented and inclusive development goal emphasized by the new urbanization strategy remains to be tested. This paper reviews the basic concept, functions and technical framework of smart community. The research on the four pilot communities in Jiangbei District shows that the decision-making of community affairs based on the platform has eliminated the dark box operation, enhanced the public trust of residents’ voting, stimulated the enthusiasm of residents to participate in community public affairs, and formed a new form of online consultation and autonomy of residents. The authors then discuss the prospects of smart community for inclusive development. This technology not only improves the efficiency of community governance, but also provides a platform for all residents to participate equally in community public affairs, reflecting the advantages of smart community construction in achieving inclusive development.
{"title":"Smart Community Building Contribute to the Realization of Inclusive Development: Cases from Jiangbei District, Nanjing, China","authors":"Zihan Xie, Dongquan Li","doi":"10.1109/UV56588.2022.10185481","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185481","url":null,"abstract":"With the development of the times, information technology is used to build smart communities, which is considered to be the way to improve community governance ability and meet the growing needs of residents for a better life. However, whether the technological revolution can achieve the people-oriented and inclusive development goal emphasized by the new urbanization strategy remains to be tested. This paper reviews the basic concept, functions and technical framework of smart community. The research on the four pilot communities in Jiangbei District shows that the decision-making of community affairs based on the platform has eliminated the dark box operation, enhanced the public trust of residents’ voting, stimulated the enthusiasm of residents to participate in community public affairs, and formed a new form of online consultation and autonomy of residents. The authors then discuss the prospects of smart community for inclusive development. This technology not only improves the efficiency of community governance, but also provides a platform for all residents to participate equally in community public affairs, reflecting the advantages of smart community construction in achieving inclusive development.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"56 5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116478462","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}
Deep learning-based detection of marine microalgae in natural waters can meet the need for rapid monitoring, facilitating researchers in marine and environmental sciences, while also paving the way for downstream cellular analysis tasks. We use a new training scheme for marine microalgae detection that consists of two phases: a teacher benchmark model phase and a student model learning phase. Using teacher model supervision to get better student model training results. Through a simple and fast image fusion method, we can obtain more realistic algae-generated images to extend the training set and eventually improve the convergence speed and performance of the model. Based on the algorithms of YOLOv5 and YOLOv6, we use the DHLC backbone network fusion method to fuse features from different levels of C3 modules and BepC3 modules together as the input of the PANet middle layer. We also use the module in BoTNet network to obtain stronger feature extraction capability by introducing self-attention mechanism in the yolo model. Since there are many small targets in marine microalgae images, we also extend the YOLOv6l model to the more powerful YOLOv6l-P6 model, which can get better detection results in the input image size of 1280. In addition, we also use time-test augmentation (TTA), weighted boxes fusion (WBF) and Single-class wighted boxes fusion (SinWBF) techniques to optimize the performance of each class. These strategies greatly improve the model detection performance and robustness under the conditions of small amount of marine microalgae microscopic image data. Finally our solution won the first place on the “Vision Meets Algae” Object Detection Challenge, and got 58.25 MAP.
{"title":"Bag of Strategies Set New State-of-the-art for Algae Object Detectors","authors":"Zhiqiang Yang, Haiming Wen, Zihan Wei, Zehan Zhang","doi":"10.1109/UV56588.2022.10185474","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185474","url":null,"abstract":"Deep learning-based detection of marine microalgae in natural waters can meet the need for rapid monitoring, facilitating researchers in marine and environmental sciences, while also paving the way for downstream cellular analysis tasks. We use a new training scheme for marine microalgae detection that consists of two phases: a teacher benchmark model phase and a student model learning phase. Using teacher model supervision to get better student model training results. Through a simple and fast image fusion method, we can obtain more realistic algae-generated images to extend the training set and eventually improve the convergence speed and performance of the model. Based on the algorithms of YOLOv5 and YOLOv6, we use the DHLC backbone network fusion method to fuse features from different levels of C3 modules and BepC3 modules together as the input of the PANet middle layer. We also use the module in BoTNet network to obtain stronger feature extraction capability by introducing self-attention mechanism in the yolo model. Since there are many small targets in marine microalgae images, we also extend the YOLOv6l model to the more powerful YOLOv6l-P6 model, which can get better detection results in the input image size of 1280. In addition, we also use time-test augmentation (TTA), weighted boxes fusion (WBF) and Single-class wighted boxes fusion (SinWBF) techniques to optimize the performance of each class. These strategies greatly improve the model detection performance and robustness under the conditions of small amount of marine microalgae microscopic image data. Finally our solution won the first place on the “Vision Meets Algae” Object Detection Challenge, and got 58.25 MAP.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"116 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131880700","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185486
Lejun Feng, H. Bai, Wenhui Shi
As a cutting-edge technology in the energy field, distributed energy systems have greater advantages over traditional energy supply models in terms of energy conservation, economy and carbon emissions. In the face of multi-type, multi-climate region and hourly fluctuating load demands, reasonable system integration design and variable working condition regulation are the keys to improving system performance. In this paper, the medium temperature heat storage unit is used as the main control method of the system, the system configuration after the system is coupled with the ORC unit is constructed, the essential difference between active energy storage and traditional passive energy storage control is explained, and the two different supply and demand of power generation excess and shortage are quantitatively analyzed. Energy storage active decoupling mechanism and active regulation method in matching scenarios.
{"title":"Research on the Regulation Mechanism of Active Energy Storage in Distributed Energy System","authors":"Lejun Feng, H. Bai, Wenhui Shi","doi":"10.1109/UV56588.2022.10185486","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185486","url":null,"abstract":"As a cutting-edge technology in the energy field, distributed energy systems have greater advantages over traditional energy supply models in terms of energy conservation, economy and carbon emissions. In the face of multi-type, multi-climate region and hourly fluctuating load demands, reasonable system integration design and variable working condition regulation are the keys to improving system performance. In this paper, the medium temperature heat storage unit is used as the main control method of the system, the system configuration after the system is coupled with the ORC unit is constructed, the essential difference between active energy storage and traditional passive energy storage control is explained, and the two different supply and demand of power generation excess and shortage are quantitatively analyzed. Energy storage active decoupling mechanism and active regulation method in matching scenarios.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128436027","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185497
Kelly Zhang, Hao Yuan, Yajun Fang
At present, the world is being faced with the challenge of an aging population, correlating to a growing number of seniors with dementia. With this uptick in persons with dementia (PWD), managing dementia-induced agitation, a behavior present in 90 precent of PWD characterized by physical aggression, verbal outburst, or other troubling behavior, is a pressing issue [1]. Caregiver burden associated with agitation is one of the leading causes a community-based PWD is institutionalized [2], so by supporting PWD-caregiver dyads, we improve individual quality of life and relieve stress placed on the global healthcare system. Use of AI technology, big data, and integrated networks of wearable and ambient sensors has enabled continuous monitoring of dementia care. However, most methods focus on data collection at the early stages of dementia. More research is needed on how novel technologies can empower PWD and their caregivers to take action to manage agitation and support them in the long term as symptoms progress. Moreover, current methods have not taken full advantage of the information obtained and do not provide personalized care. In this paper, we use the Universal Village (UV) perspective to evaluate the current status of smart technologies with the potential for use in preventing and mitigating agitation while providing support to the caregiver. We conduct evaluations based on the framework of a closed feedback control loop: data acquisition, communication, decision making, and action. We propose that a robust PWD agitation management system should take into consideration the interaction between the smart healthcare system and other seven smart city subsystems: smart home, intelligent transportation, urban planning and crowd management, smart energy management, smart city infrastructure, smart response system for city emergency, smart environmental protection and smart humanity, and also study how managing agitation would be affected by four major impacting factors of smart cities: information flow, material cycle, lifestyle, and community. This systematic study will help us explore in depth the complicated dynamic relationship between multiple impacting factors and propose a UV-oriented, integrated, resilient, inclusive, and sustainable development framework design. As such, the novel framework will improve PWD quality of life and reduce the care burden for formal and informal caregivers through continuous, unobtrusive monitoring, life-long agitation management throughout different stages of dementia, PWD-caregiver dyad-specific guidance, preventive healthcare, and timely treatment.
{"title":"Evaluation of Smart Agitation Prediction and Management for Dementia Care and Novel Universal Village Oriented Solution for Integration, Resilience, Inclusiveness and Sustainability","authors":"Kelly Zhang, Hao Yuan, Yajun Fang","doi":"10.1109/UV56588.2022.10185497","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185497","url":null,"abstract":"At present, the world is being faced with the challenge of an aging population, correlating to a growing number of seniors with dementia. With this uptick in persons with dementia (PWD), managing dementia-induced agitation, a behavior present in 90 precent of PWD characterized by physical aggression, verbal outburst, or other troubling behavior, is a pressing issue [1]. Caregiver burden associated with agitation is one of the leading causes a community-based PWD is institutionalized [2], so by supporting PWD-caregiver dyads, we improve individual quality of life and relieve stress placed on the global healthcare system. Use of AI technology, big data, and integrated networks of wearable and ambient sensors has enabled continuous monitoring of dementia care. However, most methods focus on data collection at the early stages of dementia. More research is needed on how novel technologies can empower PWD and their caregivers to take action to manage agitation and support them in the long term as symptoms progress. Moreover, current methods have not taken full advantage of the information obtained and do not provide personalized care. In this paper, we use the Universal Village (UV) perspective to evaluate the current status of smart technologies with the potential for use in preventing and mitigating agitation while providing support to the caregiver. We conduct evaluations based on the framework of a closed feedback control loop: data acquisition, communication, decision making, and action. We propose that a robust PWD agitation management system should take into consideration the interaction between the smart healthcare system and other seven smart city subsystems: smart home, intelligent transportation, urban planning and crowd management, smart energy management, smart city infrastructure, smart response system for city emergency, smart environmental protection and smart humanity, and also study how managing agitation would be affected by four major impacting factors of smart cities: information flow, material cycle, lifestyle, and community. This systematic study will help us explore in depth the complicated dynamic relationship between multiple impacting factors and propose a UV-oriented, integrated, resilient, inclusive, and sustainable development framework design. As such, the novel framework will improve PWD quality of life and reduce the care burden for formal and informal caregivers through continuous, unobtrusive monitoring, life-long agitation management throughout different stages of dementia, PWD-caregiver dyad-specific guidance, preventive healthcare, and timely treatment.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"81 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130225421","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185444
Weifan Wu, Wei Ke, Hao Sheng
Current unsupervised re-identification methods use a clustering-based neural network for training. In the vehicle re-identification field, the feature information between different vehicles is small, and it is not easy to distinguish the detailed features of different vehicles using only the basic clustering algorithm for unsupervised learning. When clustering is performed, the general clustering methods inevitably put different vehicles together due to the high similarity. We propose a new re-identification method to solve these problems. This method is based on clustering and use the unsupervised learning. First, we employ the vision transformer structure as a feature extractor. The vision transformer structure can obtain more discriminative and correlated features than the general convolution. Second, we use a fine-grained clustering method to subdivide the clustered information into different vehicles. We trained our method on two open-source datasets, and finally obtained better test results without additional labeling information.
{"title":"A Transformer-based Unsupervised Clustering Method for Vehicle Re-identification","authors":"Weifan Wu, Wei Ke, Hao Sheng","doi":"10.1109/UV56588.2022.10185444","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185444","url":null,"abstract":"Current unsupervised re-identification methods use a clustering-based neural network for training. In the vehicle re-identification field, the feature information between different vehicles is small, and it is not easy to distinguish the detailed features of different vehicles using only the basic clustering algorithm for unsupervised learning. When clustering is performed, the general clustering methods inevitably put different vehicles together due to the high similarity. We propose a new re-identification method to solve these problems. This method is based on clustering and use the unsupervised learning. First, we employ the vision transformer structure as a feature extractor. The vision transformer structure can obtain more discriminative and correlated features than the general convolution. Second, we use a fine-grained clustering method to subdivide the clustered information into different vehicles. We trained our method on two open-source datasets, and finally obtained better test results without additional labeling information.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127901798","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185498
Yiran Niu, Zhenyang Zhang, Qianling Shui
In the post-epidemic era, vaccination has become an important measure to protect the general public. In this paper, we use an ARIMA model to predict the daily number of vaccinations nationwide for the next three months by analyzing data on vaccination rates as well as the number of inhabitants, taking into account a variety of practical factors, in conjunction with the current state of the times. Indicators are rationally established, and the distribution problem is transformed into a problem of evaluating the importance of each indicator, using a simulated annealing algorithm to solve for vaccine distribution ratios for cities, neighborhoods, and towns, and to provide a reasonable vaccine distribution plan, as detailed in the model description.
{"title":"Vaccine Rational Distribution Program","authors":"Yiran Niu, Zhenyang Zhang, Qianling Shui","doi":"10.1109/UV56588.2022.10185498","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185498","url":null,"abstract":"In the post-epidemic era, vaccination has become an important measure to protect the general public. In this paper, we use an ARIMA model to predict the daily number of vaccinations nationwide for the next three months by analyzing data on vaccination rates as well as the number of inhabitants, taking into account a variety of practical factors, in conjunction with the current state of the times. Indicators are rationally established, and the distribution problem is transformed into a problem of evaluating the importance of each indicator, using a simulated annealing algorithm to solve for vaccine distribution ratios for cities, neighborhoods, and towns, and to provide a reasonable vaccine distribution plan, as detailed in the model description.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"41 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121369026","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185509
Jiaxuan Li, Muxuanzi He, P. Pang, C. Lam
As one of the essential parts of the smart city concept, the realization of intelligent transportation has become a trendy topic, and one of the fundamental ways to realize intelligent transportation is to introduce intelligent street lights. This paper mainly presents a design of a rear-mountable intelligent street light which can be installed on light posts easily. On one hand, based on the original street lights, it can fulfill all the functions of a street light after installation without the costs of replacing existing street lights. On the other hand, our approach can eliminate a lot of disassembly and installation procedures, which typically involve many engineering costs. On top of the functions provided by traditional street lights, we propose to include a AI-supported camera for traffic and parking monitoring, a full-coverage Wi-Fi access point, a laser sensor for intelligent monitoring of pedestrians crossing the road, and an electric car charging point. The co-development of the companion app can operate selected functions, for example, parking reservation and electric car charging, of the intelligent street lights. This proposal has been tested in a controlled lab environment which shows the feasibility of hardware selection, welding approaches, and the companion app design. Our future work aims to test our proposal against real-world environments and actual road conditions.
{"title":"An Easy-to-install Rear-Mountable Intelligent Street Light with Companion App for Mitigations of Urban Traffic Problems","authors":"Jiaxuan Li, Muxuanzi He, P. Pang, C. Lam","doi":"10.1109/UV56588.2022.10185509","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185509","url":null,"abstract":"As one of the essential parts of the smart city concept, the realization of intelligent transportation has become a trendy topic, and one of the fundamental ways to realize intelligent transportation is to introduce intelligent street lights. This paper mainly presents a design of a rear-mountable intelligent street light which can be installed on light posts easily. On one hand, based on the original street lights, it can fulfill all the functions of a street light after installation without the costs of replacing existing street lights. On the other hand, our approach can eliminate a lot of disassembly and installation procedures, which typically involve many engineering costs. On top of the functions provided by traditional street lights, we propose to include a AI-supported camera for traffic and parking monitoring, a full-coverage Wi-Fi access point, a laser sensor for intelligent monitoring of pedestrians crossing the road, and an electric car charging point. The co-development of the companion app can operate selected functions, for example, parking reservation and electric car charging, of the intelligent street lights. This proposal has been tested in a controlled lab environment which shows the feasibility of hardware selection, welding approaches, and the companion app design. Our future work aims to test our proposal against real-world environments and actual road conditions.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"9 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128491082","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185531
Guoyu Yang, Siyu Cheng, Jie Lei
Marine microalgae are one of the significant biological resources in marine ecosystems and a part of the “blue carbon sink.” Artificial identification of marine microalgae usually takes a lot of time, so using the object detection method to detect microalgae automatically can save a lot of artificial resources. The official website provides an algae dataset in the IEEE UV 2022 “Vision Meets Algae” object detection challenge. However, this dataset contains many small objects, which is unfavorable for the object detection model to identify algae. We use Cascade R-CNN with the backbone ConvNeXt-B as our main object detection model in this challenge. To make the model recognize small objects well, we increase the input image size and add global context to the model. During training, we used data augmentation and multi-scale training strategies that improved the performance of the model. Finally, to improve the detection performance, we integrate Cascade R-CNN, TOOD, and GFL. We evaluated our method on the test set. The mAP of Cascade R-CNN reached 54.69, while the mAP of model integration reached 56.22.
{"title":"Microalgae Detection Based on Cascade R-CNN Object Detection Model","authors":"Guoyu Yang, Siyu Cheng, Jie Lei","doi":"10.1109/UV56588.2022.10185531","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185531","url":null,"abstract":"Marine microalgae are one of the significant biological resources in marine ecosystems and a part of the “blue carbon sink.” Artificial identification of marine microalgae usually takes a lot of time, so using the object detection method to detect microalgae automatically can save a lot of artificial resources. The official website provides an algae dataset in the IEEE UV 2022 “Vision Meets Algae” object detection challenge. However, this dataset contains many small objects, which is unfavorable for the object detection model to identify algae. We use Cascade R-CNN with the backbone ConvNeXt-B as our main object detection model in this challenge. To make the model recognize small objects well, we increase the input image size and add global context to the model. During training, we used data augmentation and multi-scale training strategies that improved the performance of the model. Finally, to improve the detection performance, we integrate Cascade R-CNN, TOOD, and GFL. We evaluated our method on the test set. The mAP of Cascade R-CNN reached 54.69, while the mAP of model integration reached 56.22.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"71 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126241597","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}
Pub Date : 2022-10-22DOI: 10.1109/UV56588.2022.10185494
Wuyang Zhang, Jianming Ma
This paper aims to create an image captioning novel architecture that infuses Grid and Region-based image caption transformer, ResNet, and BART language model to offer a more detail-oriented image captioning model. Conventional state-of-the-art image captioning models mainly focuses on region-based features. They rely on decent object detector architectures like Faster R-CNN to extract object-level information to describe the image’s content. Nevertheless, they cannot remove contextual information, high computational costs, and the ability to introduce in-depth external details of objects presented in the images—the replacement of conventional CNN-based detectors results in faster computation. The experiment can generate image captions comparatively fast with higher accuracy and details with contextual information.
{"title":"Image Caption Enhancement with GRIT, Portable ResNet and BART Context-Tuning","authors":"Wuyang Zhang, Jianming Ma","doi":"10.1109/UV56588.2022.10185494","DOIUrl":"https://doi.org/10.1109/UV56588.2022.10185494","url":null,"abstract":"This paper aims to create an image captioning novel architecture that infuses Grid and Region-based image caption transformer, ResNet, and BART language model to offer a more detail-oriented image captioning model. Conventional state-of-the-art image captioning models mainly focuses on region-based features. They rely on decent object detector architectures like Faster R-CNN to extract object-level information to describe the image’s content. Nevertheless, they cannot remove contextual information, high computational costs, and the ability to introduce in-depth external details of objects presented in the images—the replacement of conventional CNN-based detectors results in faster computation. The experiment can generate image captions comparatively fast with higher accuracy and details with contextual information.","PeriodicalId":211011,"journal":{"name":"2022 6th International Conference on Universal Village (UV)","volume":"187 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114181204","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：