{"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/cleannet-transfer-learning-for-scalable-image","title":"CleanNet: Transfer Learning for Scalable Image Classifier Training with Label Noise","arxiv_id":"1711.07131","date":"2017-11-20","proceeding":"CVPR 2018 6","authors":["Kuang-Huei Lee","Xiaodong He","Lei Zhang","Linjun Yang"],"abstract":"In this paper, we study the problem of learning image classification models\nwith label noise. Existing approaches depending on human supervision are\ngenerally not scalable as manually identifying correct or incorrect labels is\ntime-consuming, whereas approaches not relying on human supervision are\nscalable but less effective. To reduce the amount of human supervision for\nlabel noise cleaning, we introduce CleanNet, a joint neural embedding network,\nwhich only requires a fraction of the classes being manually verified to\nprovide the knowledge of label noise that can be transferred to other classes.\nWe further integrate CleanNet and conventional convolutional neural network\nclassifier into one framework for image classification learning. We demonstrate\nthe effectiveness of the proposed algorithm on both of the label noise\ndetection task and the image classification on noisy data task on several\nlarge-scale datasets. Experimental results show that CleanNet can reduce label\nnoise detection error rate on held-out classes where no human supervision\navailable by 41.5% compared to current weakly supervised methods. It also\nachieves 47% of the performance gain of verifying all images with only 3.2%\nimages verified on an image classification task. Source code and dataset will\nbe available at kuanghuei.github.io/CleanNetProject.","url_abs":"http://arxiv.org/abs/1711.07131v2","url_pdf":"http://arxiv.org/pdf/1711.07131v2.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":"cleannet-transfer-learning-for-scalable-image","repo_url":"https://github.com/kuanghuei/clean-net","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"unanswered"}},{"paper_slug":"cleannet-transfer-learning-for-scalable-image","repo_url":"https://github.com/YutingLi0606/SURE","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"cleannet-transfer-learning-for-scalable-image","repo_url":"https://github.com/yingyichen-cyy/JigsawViT","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-classification-on-clothing1m-using","task":"Image Classification","dataset":"Clothing1M (using clean data)","model":"CleanNet w_soft","rank_in_archive_order":3,"of":9,"metrics":{"Accuracy":"79.90"},"uses_additional_data":true},{"leaderboard":"/sota/image-classification-on-food-101n-1","task":"Image Classification","dataset":"Food-101N","model":"CleanNet","rank_in_archive_order":2,"of":4,"metrics":{"Accuracy":"90.39"},"uses_additional_data":true}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1711.07131","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}