{"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/support-neighbor-loss-for-person-re","title":"Support Neighbor Loss for Person Re-Identification","arxiv_id":"1808.06030","date":"2018-08-18","proceeding":null,"authors":["Kai Li","Zhengming Ding","Kunpeng Li","Yulun Zhang","Yun Fu"],"abstract":"Person re-identification (re-ID) has recently been tremendously boosted due\nto the advancement of deep convolutional neural networks (CNN). The majority of\ndeep re-ID methods focus on designing new CNN architectures, while less\nattention is paid on investigating the loss functions. Verification loss and\nidentification loss are two types of losses widely used to train various deep\nre-ID models, both of which however have limitations. Verification loss guides\nthe networks to generate feature embeddings of which the intra-class variance\nis decreased while the inter-class ones is enlarged. However, training networks\nwith verification loss tends to be of slow convergence and unstable performance\nwhen the number of training samples is large. On the other hand, identification\nloss has good separating and scalable property. But its neglect to explicitly\nreduce the intra-class variance limits its performance on re-ID, because the\nsame person may have significant appearance disparity across different camera\nviews. To avoid the limitations of the two types of losses, we propose a new\nloss, called support neighbor (SN) loss. Rather than being derived from data\nsample pairs or triplets, SN loss is calculated based on the positive and\nnegative support neighbor sets of each anchor sample, which contain more\nvaluable contextual information and neighborhood structure that are beneficial\nfor more stable performance. To ensure scalability and separability, a\nsoftmax-like function is formulated to push apart the positive and negative\nsupport sets. To reduce intra-class variance, the distance between the anchor's\nnearest positive neighbor and furthest positive sample is penalized.\nIntegrating SN loss on top of Resnet50, superior re-ID results to the\nstate-of-the-art ones are obtained on several widely used datasets.","url_abs":"http://arxiv.org/abs/1808.06030v1","url_pdf":"http://arxiv.org/pdf/1808.06030v1.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":"support-neighbor-loss-for-person-re","repo_url":"https://github.com/kailigo/SN_loss_for_reID","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"person-re-identification","task_name":"Person Re-Identification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}