Jingqin Mu, Sheng Zhan, Wei Gao, Hongbo Zhang, Xianrui Deng
The determination of the type of Sanitary Wares Body (SWB) is the premise for glazing robot to intelligently choose the glazing operation mode. Traditional judgement is to sample SWB by the camera with Charge Coupled Device (CCD), however, it is necessary to build a dark room to overcome the influence of light and environment, which leads to expansion of production space and increasement of cost. For the purpose of getting over the complex circumstances, a new method for automatic extraction of the feature based on three-dimensional (3D) point cloud for SWB is put forward, which is lower demand to the light environment than two-dimensional (2D) image. There are three parts for this method: Firstly, five feature parameters of the appearance of SWB were analyzed such as length-width ratio and the number of the holes. Then after the point cloud data of SWB was captured by depth camera, preprocessing, segmentation, and projection to flat space were carried out. In the end, automatic extraction of feature parameters from grey scale image was accomplished. The experiment result showed that the parameters from point cloud were basically consistent with those from the product of sanitary wares. The approach may reduce the illumination requirement and save the production cost; Therefore, it is feasible to improve the intelligent level of ceramics process.
{"title":"Feature Analysis and Automatic Extraction for the 3D Point Cloud of the Sanitary Wares Body","authors":"Jingqin Mu, Sheng Zhan, Wei Gao, Hongbo Zhang, Xianrui Deng","doi":"10.1145/3577148.3577154","DOIUrl":"https://doi.org/10.1145/3577148.3577154","url":null,"abstract":"The determination of the type of Sanitary Wares Body (SWB) is the premise for glazing robot to intelligently choose the glazing operation mode. Traditional judgement is to sample SWB by the camera with Charge Coupled Device (CCD), however, it is necessary to build a dark room to overcome the influence of light and environment, which leads to expansion of production space and increasement of cost. For the purpose of getting over the complex circumstances, a new method for automatic extraction of the feature based on three-dimensional (3D) point cloud for SWB is put forward, which is lower demand to the light environment than two-dimensional (2D) image. There are three parts for this method: Firstly, five feature parameters of the appearance of SWB were analyzed such as length-width ratio and the number of the holes. Then after the point cloud data of SWB was captured by depth camera, preprocessing, segmentation, and projection to flat space were carried out. In the end, automatic extraction of feature parameters from grey scale image was accomplished. The experiment result showed that the parameters from point cloud were basically consistent with those from the product of sanitary wares. The approach may reduce the illumination requirement and save the production cost; Therefore, it is feasible to improve the intelligent level of ceramics process.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"25 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115501634","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}
Attention mechanism can focus on important features and suppress unnecessary features, to increase the representational power of the network, which plays an important role in visual tasks such as semantic segmentation. To this end, the UNET of Convolutional Block Attention Module (CBAM) is applied in semantic segmentation tasks to improve the performance of convolutional neural networks. The channel attention adopts maximum pooling and average pooling, and for the spatial attention, a smaller convolution kernel is proposed to reduce computation and loss of important features. Through experiments, the introduction of CBAM has a improvement in semantic segmentation tasks increasing the Validation Dice from 0.98 to 0.9871 in the Kaggle Carvana image segmentation dataset.
{"title":"Research on the Application of Unet with Convolutional Block Attention Module to Semantic Segmentation Task","authors":"Xiaotong Fang","doi":"10.1145/3577148.3577151","DOIUrl":"https://doi.org/10.1145/3577148.3577151","url":null,"abstract":"Attention mechanism can focus on important features and suppress unnecessary features, to increase the representational power of the network, which plays an important role in visual tasks such as semantic segmentation. To this end, the UNET of Convolutional Block Attention Module (CBAM) is applied in semantic segmentation tasks to improve the performance of convolutional neural networks. The channel attention adopts maximum pooling and average pooling, and for the spatial attention, a smaller convolution kernel is proposed to reduce computation and loss of important features. Through experiments, the introduction of CBAM has a improvement in semantic segmentation tasks increasing the Validation Dice from 0.98 to 0.9871 in the Kaggle Carvana image segmentation dataset.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115991526","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}
W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra
ICTS are revolutionizing not only the way of communication and interrelation between people, but also between information systems, in such a way that they allow the exchange of information between computer systems, a method will be developed to take advantage of the functionalities of the information systems dedicated to the health area, where information related to the health area can be shared, to improve the procedures related to decision making, which are vital in emergency situations, as in the case of an emergency care where it is required to know if the patient has some type of insurance or if he/she is allergic to certain medications, this information is important if an online consultation is made with the information systems. In this sense the information systems that work in the Health System, communicates with the information system of the insurance companies, to request the data of the insurance policy, as well as the hospital systems can communicate with the service providers to request medical supplies, among other applications, thanks to the interoperability, based on the XML communication standard that in its essence is the basis of the HL7 protocol, the results of the proposed methodology allow the interaction between systems, if as a development and application guide where you can create and read these messages based on XML, which can be applicable and scalable.
{"title":"Methodology for Interoperability between Health Information Systems, for Information Management and Decision-Making","authors":"W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra","doi":"10.1145/3577148.3577156","DOIUrl":"https://doi.org/10.1145/3577148.3577156","url":null,"abstract":"ICTS are revolutionizing not only the way of communication and interrelation between people, but also between information systems, in such a way that they allow the exchange of information between computer systems, a method will be developed to take advantage of the functionalities of the information systems dedicated to the health area, where information related to the health area can be shared, to improve the procedures related to decision making, which are vital in emergency situations, as in the case of an emergency care where it is required to know if the patient has some type of insurance or if he/she is allergic to certain medications, this information is important if an online consultation is made with the information systems. In this sense the information systems that work in the Health System, communicates with the information system of the insurance companies, to request the data of the insurance policy, as well as the hospital systems can communicate with the service providers to request medical supplies, among other applications, thanks to the interoperability, based on the XML communication standard that in its essence is the basis of the HL7 protocol, the results of the proposed methodology allow the interaction between systems, if as a development and application guide where you can create and read these messages based on XML, which can be applicable and scalable.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"48 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130499009","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}
With the improvement of people's living standards, the demand for poultry has further increased. The screening of cock reproductive performance according to semen quality has become one of the attention directions. It is time-consuming to screen the breeding performance of cocks based on human vision, and there will be recognition errors. In this study, combined with the hypothesis that there is a correlation between cockscomb characteristics and semen quality, a scheme based on computer vision is proposed to avoid this problem. The collected cockscomb data are input into our model, which can automatically judge the breeding performance of cocks. We use transfer learning and change the weight ratio of the latest fine-grained visual classification algorithm Transfg at different depths to improve the accuracy of our data set. The average accuracy of the model in the test set is 44.6% (the data set contains 2053 pictures in the training set and 505 pictures in the test set), which is 1% better than the original vision transformer and 1.3% better than the convolution network model.
{"title":"Judgment Model of Cock Reproductive Performance based on Vison Transformer","authors":"Xuhong Lin, Qian Yan, Caicong Wu, Yifei Chen","doi":"10.1145/3577148.3577155","DOIUrl":"https://doi.org/10.1145/3577148.3577155","url":null,"abstract":"With the improvement of people's living standards, the demand for poultry has further increased. The screening of cock reproductive performance according to semen quality has become one of the attention directions. It is time-consuming to screen the breeding performance of cocks based on human vision, and there will be recognition errors. In this study, combined with the hypothesis that there is a correlation between cockscomb characteristics and semen quality, a scheme based on computer vision is proposed to avoid this problem. The collected cockscomb data are input into our model, which can automatically judge the breeding performance of cocks. We use transfer learning and change the weight ratio of the latest fine-grained visual classification algorithm Transfg at different depths to improve the accuracy of our data set. The average accuracy of the model in the test set is 44.6% (the data set contains 2053 pictures in the training set and 505 pictures in the test set), which is 1% better than the original vision transformer and 1.3% better than the convolution network model.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"101 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131276026","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}
The research of infinite dimensional dynamic system arose in the 1980s. With Mandelbrot's fractal theory, the theory of fractional calculus, as the basic tool of fractal theory, has been widely concerned and applied, which makes the theory of fractional calculus develop rapidly. In past decades, the fractional calculus theory has been widely used in many fields. The fractional differential equation model can more accurately simulate practical problems than the integer order model, which makes the fractional differential equation become the current research hotspot. However, it is difficult for us to obtain the explicit solution of most Nonlinear Fractional Ordinary Differential Equations. Therefore, the focus of the research on Fractional Ordinary differential equations has shifted to the geometric and topological properties of solutions. As an important part of studying lattice systems, attractors are used to describe the geometric and topological properties of solutions of lattice systems. At present, the research on the solutions of most fractional order lattice systems is only limited to discussing the existence of solutions in finite intervals. However, there have been few relevant results on the existence of solutions in the whole space of fractional order lattice systems. Therefore, it is meaningful to study the existence of solutions in the whole space of fractional order Fitzhugh Nagumo lattice systems.
{"title":"Existence and Uniqueness of Solution of Fractional FitzHugh-Nagumo System","authors":"Xuyi Wu, Zhenqi Zhang","doi":"10.1145/3577148.3577158","DOIUrl":"https://doi.org/10.1145/3577148.3577158","url":null,"abstract":"The research of infinite dimensional dynamic system arose in the 1980s. With Mandelbrot's fractal theory, the theory of fractional calculus, as the basic tool of fractal theory, has been widely concerned and applied, which makes the theory of fractional calculus develop rapidly. In past decades, the fractional calculus theory has been widely used in many fields. The fractional differential equation model can more accurately simulate practical problems than the integer order model, which makes the fractional differential equation become the current research hotspot. However, it is difficult for us to obtain the explicit solution of most Nonlinear Fractional Ordinary Differential Equations. Therefore, the focus of the research on Fractional Ordinary differential equations has shifted to the geometric and topological properties of solutions. As an important part of studying lattice systems, attractors are used to describe the geometric and topological properties of solutions of lattice systems. At present, the research on the solutions of most fractional order lattice systems is only limited to discussing the existence of solutions in finite intervals. However, there have been few relevant results on the existence of solutions in the whole space of fractional order lattice systems. Therefore, it is meaningful to study the existence of solutions in the whole space of fractional order Fitzhugh Nagumo lattice systems.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116999022","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}
W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra
In the process of detection and diagnosis of sleep apnea, we find multiple complex systems that perform the evaluation of the characteristic parameters of this pathology, such as heart rate monitoring, measurement of breathing rate, the presence of snoring, leg movement, among others. Each of them allows the identification of the degree of sleep APNEA. In the present work we present a low cost vision system for monitoring body movement when the person is sleeping, the system presented allows to evaluate the person at home, without the need to be in a specialized center, the connectivity of the equipment allows to evaluate the patient remotely, as well as allows an evaluation of the video recording, only in the case that the patient presents some characteristic movement, the system only records when the patient moves. The system can be applied and scaled depending on the required conditions. Accompanying the video recording we use smart watches to evaluate the levels of sleep, oxygen saturation and heart rate, as a result we present the data of the averages obtained to synchronize with the recording.
{"title":"Low-Cost Video System for the Monitoring Process in the Detection of Sleep Apnea","authors":"W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra","doi":"10.1145/3577148.3577149","DOIUrl":"https://doi.org/10.1145/3577148.3577149","url":null,"abstract":"In the process of detection and diagnosis of sleep apnea, we find multiple complex systems that perform the evaluation of the characteristic parameters of this pathology, such as heart rate monitoring, measurement of breathing rate, the presence of snoring, leg movement, among others. Each of them allows the identification of the degree of sleep APNEA. In the present work we present a low cost vision system for monitoring body movement when the person is sleeping, the system presented allows to evaluate the person at home, without the need to be in a specialized center, the connectivity of the equipment allows to evaluate the patient remotely, as well as allows an evaluation of the video recording, only in the case that the patient presents some characteristic movement, the system only records when the patient moves. The system can be applied and scaled depending on the required conditions. Accompanying the video recording we use smart watches to evaluate the levels of sleep, oxygen saturation and heart rate, as a result we present the data of the averages obtained to synchronize with the recording.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128648293","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 explored the application of deep learning target detection methods in the field of X-ray security screening. Faster R-CNN is a fully supervised deep learning method that uses only abnormal images containing dangerous goods as the training set, thus making it difficult to learn the features of normal images. It results in its high false detection rate when detecting normal images. In view of the above problems, combined with the characteristics of most of the X-ray security images are normal images, the author proposed a pre-classified head X-ray security image recognition method to reduce the false detection rate, while improving the performance and efficiency of dangerous goods detection, and more suitable for real X-ray security application scenarios.
{"title":"Research on the Identification Method of Dangerous Goods in Security Inspection Images Based on Deep Learning","authors":"Yuan-Fang Li","doi":"10.1145/3577148.3577153","DOIUrl":"https://doi.org/10.1145/3577148.3577153","url":null,"abstract":"This paper explored the application of deep learning target detection methods in the field of X-ray security screening. Faster R-CNN is a fully supervised deep learning method that uses only abnormal images containing dangerous goods as the training set, thus making it difficult to learn the features of normal images. It results in its high false detection rate when detecting normal images. In view of the above problems, combined with the characteristics of most of the X-ray security images are normal images, the author proposed a pre-classified head X-ray security image recognition method to reduce the false detection rate, while improving the performance and efficiency of dangerous goods detection, and more suitable for real X-ray security application scenarios.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"11 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131494487","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 is an exploration of the recognition technology of pneumonia pictures based on convolutional neural network technology. Among them, the recognition model used is based on the study of hundreds of real X-rays of lung pictures database, which contains not only lung pictures of real pneumonia patients, but also lung pictures of normal people. This article describes the most popular techniques of the moment, convolutional neural networks, which are widely used in areas such as image recognition or machine learning and are recognized by most people. This paper analyzes the specific implementation techniques of convolutional neural networks used, and uses some new methods to optimize and implement this algorithm, so as to achieve a better model structure and accuracy. Among them, with regard to the pooling layer, working between the convolutional layer and the final output layer, this paper compares various pooling methods and finally yields the maximum pooled neural network is more stable, and the average pooled neural network is more effective for large databases. The final use, pooling the resulting model accuracy is about 95% by maximum pooled neural network.
{"title":"Pneumonia image classification method based on improved convolutional neural network","authors":"Yuyang Tan, Toe Teoh Teik","doi":"10.1145/3577148.3577150","DOIUrl":"https://doi.org/10.1145/3577148.3577150","url":null,"abstract":"This is an exploration of the recognition technology of pneumonia pictures based on convolutional neural network technology. Among them, the recognition model used is based on the study of hundreds of real X-rays of lung pictures database, which contains not only lung pictures of real pneumonia patients, but also lung pictures of normal people. This article describes the most popular techniques of the moment, convolutional neural networks, which are widely used in areas such as image recognition or machine learning and are recognized by most people. This paper analyzes the specific implementation techniques of convolutional neural networks used, and uses some new methods to optimize and implement this algorithm, so as to achieve a better model structure and accuracy. Among them, with regard to the pooling layer, working between the convolutional layer and the final output layer, this paper compares various pooling methods and finally yields the maximum pooled neural network is more stable, and the average pooled neural network is more effective for large databases. The final use, pooling the resulting model accuracy is about 95% by maximum pooled neural network.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123729013","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}
W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra
One of the effects of the COVID-19 pandemic is the adaptation of most of the activities remotely or virtually, in the case of medical appointments, in the different specialties other than emergencies produced by COVID-19. Most of them continued in the format through online appointments. One of the important processes in medical evaluations is related to the so-called specialist boards, where special cases are evaluated, for which several physicians must be connected online, in addition to being able to make reports jointly. In this paper we develop a methodology to perform medical meetings of specialists, using a platform dedicated to the use in video games, through the DISCORD tool interconnectivity from various devices is performed, the results demonstrate the interactivity and applicability of the methodology, so it can be applied in different processes where interconnectivity between different devices and the concurrence of several users is required. We present a methodology to configure virtual appointment rooms, the results allow us to verify that the methodology can be replicated and scaled according to the needs.
{"title":"Methodology to carry out medical meeting online, based on low-cost solutions","authors":"W. Auccahuasi, Sandra Meza, K. Rojas, Oscar Linares, Miryam Inciso-Rojas, Aly Auccahuasi, Edward Flores, E. Felix, J. Aybar, Tamara Pando-Ezcurra","doi":"10.1145/3577148.3577162","DOIUrl":"https://doi.org/10.1145/3577148.3577162","url":null,"abstract":"One of the effects of the COVID-19 pandemic is the adaptation of most of the activities remotely or virtually, in the case of medical appointments, in the different specialties other than emergencies produced by COVID-19. Most of them continued in the format through online appointments. One of the important processes in medical evaluations is related to the so-called specialist boards, where special cases are evaluated, for which several physicians must be connected online, in addition to being able to make reports jointly. In this paper we develop a methodology to perform medical meetings of specialists, using a platform dedicated to the use in video games, through the DISCORD tool interconnectivity from various devices is performed, the results demonstrate the interactivity and applicability of the methodology, so it can be applied in different processes where interconnectivity between different devices and the concurrence of several users is required. We present a methodology to configure virtual appointment rooms, the results allow us to verify that the methodology can be replicated and scaled according to the needs.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133125926","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 image restoration techniques have made great progress in recent years, and EdgeConnect network has achieved good results in image restoration. We find that EdgeConnect suffers from a complex training process and poor migratability, which reduces its usability in practical applications. We explore the reasons for the poor transferability learning and generalization of the EdgeConnect model, and propose a small-sample unsupervised joint transfer learning method for the case of small datasets and low data similarity. The method combines a large amount of Fine-tune with a small amount of direct migration training to enable the network to learn new knowledge of the target domain while avoiding overfitting and negative migration. We perform migration learning and evaluation on 600 images from Paris StreetView with a pre-trained model obtained on the CelebA dataset, and show that it outperforms other current methods in terms of quality.
{"title":"Unsupervised Transfer Learning for Generative Image Inpainting with Adversarial Edge Learning","authors":"Yiming Zhao, Yuxiang Zhang, Zishuo Sun","doi":"10.1145/3577148.3577152","DOIUrl":"https://doi.org/10.1145/3577148.3577152","url":null,"abstract":"Deep learning-based image restoration techniques have made great progress in recent years, and EdgeConnect network has achieved good results in image restoration. We find that EdgeConnect suffers from a complex training process and poor migratability, which reduces its usability in practical applications. We explore the reasons for the poor transferability learning and generalization of the EdgeConnect model, and propose a small-sample unsupervised joint transfer learning method for the case of small datasets and low data similarity. The method combines a large amount of Fine-tune with a small amount of direct migration training to enable the network to learn new knowledge of the target domain while avoiding overfitting and negative migration. We perform migration learning and evaluation on 600 images from Paris StreetView with a pre-trained model obtained on the CelebA dataset, and show that it outperforms other current methods in terms of quality.","PeriodicalId":107500,"journal":{"name":"Proceedings of the 2022 5th International Conference on Sensors, Signal and Image Processing","volume":"87 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126177032","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：