{"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/deep-label-distribution-learning-with-label","title":"Deep Label Distribution Learning with Label Ambiguity","arxiv_id":"1611.01731","date":"2016-11-06","proceeding":null,"authors":["Bin-Bin Gao","Chao Xing","Chen-Wei Xie","Jianxin Wu","Xin Geng"],"abstract":"Convolutional Neural Networks (ConvNets) have achieved excellent recognition\nperformance in various visual recognition tasks. A large labeled training set\nis one of the most important factors for its success. However, it is difficult\nto collect sufficient training images with precise labels in some domains such\nas apparent age estimation, head pose estimation, multi-label classification\nand semantic segmentation. Fortunately, there is ambiguous information among\nlabels, which makes these tasks different from traditional classification.\nBased on this observation, we convert the label of each image into a discrete\nlabel distribution, and learn the label distribution by minimizing a\nKullback-Leibler divergence between the predicted and ground-truth label\ndistributions using deep ConvNets. The proposed DLDL (Deep Label Distribution\nLearning) method effectively utilizes the label ambiguity in both feature\nlearning and classifier learning, which help prevent the network from\nover-fitting even when the training set is small. Experimental results show\nthat the proposed approach produces significantly better results than\nstate-of-the-art methods for age estimation and head pose estimation. At the\nsame time, it also improves recognition performance for multi-label\nclassification and semantic segmentation tasks.","url_abs":"http://arxiv.org/abs/1611.01731v2","url_pdf":"http://arxiv.org/pdf/1611.01731v2.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":"deep-label-distribution-learning-with-label","repo_url":"https://github.com/gaobb/DLDL","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok"}},{"paper_slug":"deep-label-distribution-learning-with-label","repo_url":"https://github.com/paplhjak/facial-age-estimation-benchmark","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"age-estimation","task_name":"Age Estimation"},{"task_slug":"classification-1","task_name":"Classification"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"head-pose-estimation","task_name":"Head Pose Estimation"},{"task_slug":"multi-label-classification-2","task_name":"MUlTI-LABEL-ClASSIFICATION"},{"task_slug":"multi-label-classification","task_name":"Multi-Label Classification"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"},{"task_slug":"semantic-segmentation","task_name":"Semantic Segmentation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/age-estimation-on-chalearn-2015","task":"Age Estimation","dataset":"ChaLearn 2015","model":"DLDL+VGG-Face","rank_in_archive_order":6,"of":7,"metrics":{"MAE":"3.51","e-error":" 0.31"},"uses_additional_data":false},{"leaderboard":"/sota/age-estimation-on-morph-album2","task":"Age Estimation","dataset":"MORPH Album2","model":"DLDL+VGG-Face","rank_in_archive_order":5,"of":9,"metrics":{"MAE":"2.42±0.01"},"uses_additional_data":false},{"leaderboard":"/sota/age-estimation-on-morph-album2","task":"Age Estimation","dataset":"MORPH Album2","model":"DLDL+VGG-Face (KL, Max)3","rank_in_archive_order":6,"of":9,"metrics":{"MAE":"2.42"},"uses_additional_data":false},{"leaderboard":"/sota/head-pose-estimation-on-aflw","task":"Head Pose Estimation","dataset":"AFLW","model":"DLDL (KL)","rank_in_archive_order":6,"of":6,"metrics":{"MAE":"9.78"},"uses_additional_data":false},{"leaderboard":"/sota/head-pose-estimation-on-bjut-3d","task":"Head Pose Estimation","dataset":"BJUT-3D","model":"Ours DLDL (KL)","rank_in_archive_order":1,"of":1,"metrics":{"MAE":"0.09"},"uses_additional_data":false},{"leaderboard":"/sota/head-pose-estimation-on-pointing-04","task":"Head Pose Estimation","dataset":"Pointing'04","model":"Ours DLDL (KL)","rank_in_archive_order":1,"of":1,"metrics":{"MAE":"4.64"},"uses_additional_data":false},{"leaderboard":"/sota/multi-label-classification-on-pascal-voc-2007","task":"Multi-Label Classification","dataset":"PASCAL VOC 2007","model":"Ours PF-DLDL","rank_in_archive_order":15,"of":17,"metrics":{"mAP":"93.4"},"uses_additional_data":false},{"leaderboard":"/sota/multi-label-classification-on-pascal-voc-2012","task":"Multi-Label Classification","dataset":"PASCAL VOC 2012","model":"Ours PF-DLDL","rank_in_archive_order":3,"of":3,"metrics":{"mAP":"92.4"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-segmentation-on-pascal-voc-2011","task":"Semantic Segmentation","dataset":"PASCAL VOC 2011","model":"DLDL-8s+CRF","rank_in_archive_order":1,"of":1,"metrics":{"Mean IoU":"67.6"},"uses_additional_data":false},{"leaderboard":"/sota/semantic-segmentation-on-pascal-voc-2012-1","task":"Semantic Segmentation","dataset":"PASCAL VOC 2012","model":"DLDL-8s+CRF","rank_in_archive_order":1,"of":1,"metrics":{"Mean IoU":"67.1"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1611.01731","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}