The applications of machine learning(ML)in the medical domain are often hindered by the limited availability of high-quality data.To address this challenge,we explore the synthetic generation of echocardiography image...The applications of machine learning(ML)in the medical domain are often hindered by the limited availability of high-quality data.To address this challenge,we explore the synthetic generation of echocardiography images(echoCG)using state-of-the-art generative models.We conduct a comprehensive evaluation of three prominent methods:Cycle-consistent generative adversarial network(CycleGAN),Contrastive Unpaired Translation(CUT),and Stable Diffusion 1.5 with Low-Rank Adaptation(LoRA).Our research presents the data generation methodol-ogy,image samples,and evaluation strategy,followed by an extensive user study involving licensed cardiologists and surgeons who assess the perceived quality and medical soundness of the generated images.Our findings indicate that Stable Diffusion outperforms both CycleGAN and CUT in generating images that are nearly indistinguishable from real echoCG images,making it a promising tool for augmenting medical datasets.However,we also identify limitations in the synthetic images generated by CycleGAN and CUT,which are easily distinguishable as non-realistic by medical professionals.This study highlights the potential of diffusion models in medical imaging and their applicability in addressing data scarcity,while also outlining the areas for future improvement.展开更多
In pedestrian re-recognition,the traditional pedestrian re-recognition method will be affected by the changes of background,veil,clothing and so on,which will make the recognition effect decline.In order to reduce the...In pedestrian re-recognition,the traditional pedestrian re-recognition method will be affected by the changes of background,veil,clothing and so on,which will make the recognition effect decline.In order to reduce the impact of background,veil,clothing and other changes on the recognition effect,this paper proposes a pedestrian re-recognition method based on the cycle-consistent generative adversarial network and multifeature fusion.By comparing the measured distance between two pedestrians,pedestrian re-recognition is accomplished.Firstly,this paper uses Cycle GAN to transform and expand the data set,so as to reduce the influence of pedestrian posture changes as much as possible.The method consists of two branches:global feature extraction and local feature extraction.Then the global feature and local feature are fused.The fused features are used for comparison measurement learning,and the similarity scores are calculated to sort the samples.A large number of experimental results on large data sets CUHK03 and VIPER show that this new method reduces the influence of background,veil,clothing and other changes on the recognition effect.展开更多
Side-scan sonar(SSS)is essential for acquiring high-resolution seafloor images over large areas,facilitat-ing the identification of subsea objects.However,military security restrictions and the scarcity of subsea targ...Side-scan sonar(SSS)is essential for acquiring high-resolution seafloor images over large areas,facilitat-ing the identification of subsea objects.However,military security restrictions and the scarcity of subsea targets limit the availability of SSS data,posing challenges for Automatic Target Recognition(ATR)research.This paper presents an approach that uses Cycle-Consistent Generative Adversarial Networks(CycleGAN)to augment SSS images of key subsea objects,such as shipwrecks,aircraft,and drowning victims.The process begins by constructing 3D models to generate rendered images with realistic shadows frommultiple angles.To enhance image quality,a shadowextractor and shadow region loss function are introduced to ensure consistent shadowrepresentation.Additionally,amulti-resolution learning structure enables effective training,even with limited data availability.The experimental results show that the generated data improved object detection accuracy when they were used for training and demonstrated the ability to generate clear shadow and background regions with stability.展开更多
As the 6G era approaches,wireless communication faces challenges such as massive user numbers,high mobility,and spectrum resource sharing.Radio maps are crucial for network design,optimization,and management,providing...As the 6G era approaches,wireless communication faces challenges such as massive user numbers,high mobility,and spectrum resource sharing.Radio maps are crucial for network design,optimization,and management,providing essential channel information.In this paper,we propose an innovative learning framework for Radio Map Estimation(RME)based on cycle-consistent generative adversarial networks.Traditional RME methods are often constrained by model complexity and interpolation accuracy,while learning-based methods require strictly paired datasets,making their practical application difficult.Our method overcomes these limitations by enabling training with unpaired data,efficiently converting local features into radio maps.Our experimental results demonstrate the effectiveness of the proposed method in two scenarios:accurate map data and map data with dynamic errors.To address dynamic interference,we designed a two-stage learning process that uses sparse observations to correct local details in the radio map,and the model's accuracy and practicality.展开更多
基金funded by the Science Committee of the Ministry of Science and Higher Education of the Republic of Kazakhstan(Grant No.AP13068032-Development of Methods and Algorithms for Machine Learning for Predicting Pathologies of the Cardiovascular System Based on Echocardiography and Electrocardiography).
文摘The applications of machine learning(ML)in the medical domain are often hindered by the limited availability of high-quality data.To address this challenge,we explore the synthetic generation of echocardiography images(echoCG)using state-of-the-art generative models.We conduct a comprehensive evaluation of three prominent methods:Cycle-consistent generative adversarial network(CycleGAN),Contrastive Unpaired Translation(CUT),and Stable Diffusion 1.5 with Low-Rank Adaptation(LoRA).Our research presents the data generation methodol-ogy,image samples,and evaluation strategy,followed by an extensive user study involving licensed cardiologists and surgeons who assess the perceived quality and medical soundness of the generated images.Our findings indicate that Stable Diffusion outperforms both CycleGAN and CUT in generating images that are nearly indistinguishable from real echoCG images,making it a promising tool for augmenting medical datasets.However,we also identify limitations in the synthetic images generated by CycleGAN and CUT,which are easily distinguishable as non-realistic by medical professionals.This study highlights the potential of diffusion models in medical imaging and their applicability in addressing data scarcity,while also outlining the areas for future improvement.
文摘In pedestrian re-recognition,the traditional pedestrian re-recognition method will be affected by the changes of background,veil,clothing and so on,which will make the recognition effect decline.In order to reduce the impact of background,veil,clothing and other changes on the recognition effect,this paper proposes a pedestrian re-recognition method based on the cycle-consistent generative adversarial network and multifeature fusion.By comparing the measured distance between two pedestrians,pedestrian re-recognition is accomplished.Firstly,this paper uses Cycle GAN to transform and expand the data set,so as to reduce the influence of pedestrian posture changes as much as possible.The method consists of two branches:global feature extraction and local feature extraction.Then the global feature and local feature are fused.The fused features are used for comparison measurement learning,and the similarity scores are calculated to sort the samples.A large number of experimental results on large data sets CUHK03 and VIPER show that this new method reduces the influence of background,veil,clothing and other changes on the recognition effect.
基金supported by the National Research Foundation of Korea(NRF)grant funded by the Korea government(MSIT)(No.RS-2024-00334159)the Korea Institute of Ocean Science and Technology(KIOST)project entitled“Development of Maritime Domain Awareness Technology for Sea Power Enhancement”(PEA0332).
文摘Side-scan sonar(SSS)is essential for acquiring high-resolution seafloor images over large areas,facilitat-ing the identification of subsea objects.However,military security restrictions and the scarcity of subsea targets limit the availability of SSS data,posing challenges for Automatic Target Recognition(ATR)research.This paper presents an approach that uses Cycle-Consistent Generative Adversarial Networks(CycleGAN)to augment SSS images of key subsea objects,such as shipwrecks,aircraft,and drowning victims.The process begins by constructing 3D models to generate rendered images with realistic shadows frommultiple angles.To enhance image quality,a shadowextractor and shadow region loss function are introduced to ensure consistent shadowrepresentation.Additionally,amulti-resolution learning structure enables effective training,even with limited data availability.The experimental results show that the generated data improved object detection accuracy when they were used for training and demonstrated the ability to generate clear shadow and background regions with stability.
基金supported in part by the Shenzhen Basic Research Program under Grant JCYJ20220531103008018,and Grants 20231120142345001 and 20231127144045001the Guangdong Basic Research Program under Grant 2024ZDZX1016。
文摘As the 6G era approaches,wireless communication faces challenges such as massive user numbers,high mobility,and spectrum resource sharing.Radio maps are crucial for network design,optimization,and management,providing essential channel information.In this paper,we propose an innovative learning framework for Radio Map Estimation(RME)based on cycle-consistent generative adversarial networks.Traditional RME methods are often constrained by model complexity and interpolation accuracy,while learning-based methods require strictly paired datasets,making their practical application difficult.Our method overcomes these limitations by enabling training with unpaired data,efficiently converting local features into radio maps.Our experimental results demonstrate the effectiveness of the proposed method in two scenarios:accurate map data and map data with dynamic errors.To address dynamic interference,we designed a two-stage learning process that uses sparse observations to correct local details in the radio map,and the model's accuracy and practicality.
