{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/unsupervised-person-re-identification","title":"Unsupervised Person Re-identification: Clustering and Fine-tuning","arxiv_id":"1705.10444","date":"2017-05-30","proceeding":null,"authors":["Hehe Fan","Liang Zheng","Yi Yang"],"abstract":"The superiority of deeply learned pedestrian representations has been\nreported in very recent literature of person re-identification (re-ID). In this\npaper, we consider the more pragmatic issue of learning a deep feature with no\nor only a few labels. We propose a progressive unsupervised learning (PUL)\nmethod to transfer pretrained deep representations to unseen domains. Our\nmethod is easy to implement and can be viewed as an effective baseline for\nunsupervised re-ID feature learning. Specifically, PUL iterates between 1)\npedestrian clustering and 2) fine-tuning of the convolutional neural network\n(CNN) to improve the original model trained on the irrelevant labeled dataset.\nSince the clustering results can be very noisy, we add a selection operation\nbetween the clustering and fine-tuning. At the beginning when the model is\nweak, CNN is fine-tuned on a small amount of reliable examples which locate\nnear to cluster centroids in the feature space. As the model becomes stronger\nin subsequent iterations, more images are being adaptively selected as CNN\ntraining samples. Progressively, pedestrian clustering and the CNN model are\nimproved simultaneously until algorithm convergence. This process is naturally\nformulated as self-paced learning. We then point out promising directions that\nmay lead to further improvement. Extensive experiments on three large-scale\nre-ID datasets demonstrate that PUL outputs discriminative features that\nimprove the re-ID accuracy.","url_abs":"http://arxiv.org/abs/1705.10444v2","url_pdf":"http://arxiv.org/pdf/1705.10444v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"unsupervised-person-re-identification","repo_url":"https://github.com/hehefan/Unsupervised-Person-Re-identification-Clustering-and-Fine-tuning","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"clustering","task_name":"Clustering"},{"task_slug":"person-re-identification","task_name":"Person Re-Identification"},{"task_slug":"unsupervised-domain-adaptation","task_name":"Unsupervised Domain Adaptation"},{"task_slug":"unsupervised-person-re-identification","task_name":"Unsupervised Person Re-Identification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/person-re-identification-on-dukemtmc-reid","task":"Person Re-Identification","dataset":"DukeMTMC-reID","model":"PUL*","rank_in_archive_order":93,"of":94,"metrics":{"Rank-1":"30.4","mAP":"16.4"},"uses_additional_data":false},{"leaderboard":"/sota/person-re-identification-on-market-1501","task":"Person Re-Identification","dataset":"Market-1501","model":"PUL*","rank_in_archive_order":123,"of":135,"metrics":{"Rank-1":"44.7","mAP":"20.1"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-domain-adaptation-on-vehicleid","task":"Unsupervised Domain Adaptation","dataset":"VehicleID to VeRi-776","model":"PUL","rank_in_archive_order":13,"of":14,"metrics":{"Rank-1":"55.24","Rank-10":"-","Rank-5":"67.34","mAP":"17.06"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-domain-adaptation-on-veri-776-to-2","task":"Unsupervised Domain Adaptation","dataset":"Veri-776 to VehicleID Large","model":"PUL","rank_in_archive_order":10,"of":13,"metrics":{"R-1":"30.90","R-10":"-","R-5":"47.18","mAP":"34.71"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-domain-adaptation-on-veri-776-to-1","task":"Unsupervised Domain Adaptation","dataset":"Veri-776 to VehicleID Medium","model":"PUL","rank_in_archive_order":10,"of":13,"metrics":{"R-1":"33.83","R-10":"-","R-5":"49.72","mAP":"37.68"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-domain-adaptation-on-veri-776-to","task":"Unsupervised Domain Adaptation","dataset":"Veri-776 to VehicleID Small","model":"PUL","rank_in_archive_order":6,"of":8,"metrics":{" mAP":"43.90","R-1":"40.03","R-10":"-","R-5":"56.03"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-person-re-identification-on-5","task":"Unsupervised Person Re-Identification","dataset":"DukeMTMC-reID","model":"PUL","rank_in_archive_order":12,"of":13,"metrics":{"MAP":"16.4","Rank-1":"30.0","Rank-10":"48.5","Rank-5":"43.4"},"uses_additional_data":false},{"leaderboard":"/sota/unsupervised-person-re-identification-on-4","task":"Unsupervised Person Re-Identification","dataset":"Market-1501","model":"PUL","rank_in_archive_order":22,"of":23,"metrics":{"MAP":"20.5","Rank-1":"45.5","Rank-10":"66.7","Rank-5":"60.7"},"uses_additional_data":false}],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}