{"title":"基于深度学习的行人再识别算法框架","authors":"Huawei Wang, Yijing Guo","doi":"10.1117/12.2653790","DOIUrl":null,"url":null,"abstract":"To further promote the improvement of pedestrian re-identification performance, this paper studies the reid framework based on \"reid-strong-baseline\", and uses different optimization schemes to improve the network performance. Firstly, the study tests three kinds of loss: Softmax, triplet hard, and Softmax + triplet hard, to verify the Rank-1 performance obtained and which can achieve the best performance. Secondly, based on the prototype network obtained by applying Softmax + triplet hard loss, we utilize several optimization methods including data enhancement, learning rate optimization, sampling method, and Label smoothing. Then we study the effectiveness of these optimizations on the performance of the Baseline model and the degree of improvement. Finally, this paper studies the efficiency of different Backbone and network depths on the performance of pedestrian re-identification.","PeriodicalId":32903,"journal":{"name":"JITeCS Journal of Information Technology and Computer Science","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2022-12-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"A novel pedestrian re-identification algorithm framework based on deep learning\",\"authors\":\"Huawei Wang, Yijing Guo\",\"doi\":\"10.1117/12.2653790\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"To further promote the improvement of pedestrian re-identification performance, this paper studies the reid framework based on \\\"reid-strong-baseline\\\", and uses different optimization schemes to improve the network performance. Firstly, the study tests three kinds of loss: Softmax, triplet hard, and Softmax + triplet hard, to verify the Rank-1 performance obtained and which can achieve the best performance. Secondly, based on the prototype network obtained by applying Softmax + triplet hard loss, we utilize several optimization methods including data enhancement, learning rate optimization, sampling method, and Label smoothing. Then we study the effectiveness of these optimizations on the performance of the Baseline model and the degree of improvement. Finally, this paper studies the efficiency of different Backbone and network depths on the performance of pedestrian re-identification.\",\"PeriodicalId\":32903,\"journal\":{\"name\":\"JITeCS Journal of Information Technology and Computer Science\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2022-12-08\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"JITeCS Journal of Information Technology and Computer Science\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1117/12.2653790\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"JITeCS Journal of Information Technology and Computer Science","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1117/12.2653790","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
A novel pedestrian re-identification algorithm framework based on deep learning
To further promote the improvement of pedestrian re-identification performance, this paper studies the reid framework based on "reid-strong-baseline", and uses different optimization schemes to improve the network performance. Firstly, the study tests three kinds of loss: Softmax, triplet hard, and Softmax + triplet hard, to verify the Rank-1 performance obtained and which can achieve the best performance. Secondly, based on the prototype network obtained by applying Softmax + triplet hard loss, we utilize several optimization methods including data enhancement, learning rate optimization, sampling method, and Label smoothing. Then we study the effectiveness of these optimizations on the performance of the Baseline model and the degree of improvement. Finally, this paper studies the efficiency of different Backbone and network depths on the performance of pedestrian re-identification.