{"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/quantized-densely-connected-u-nets-for","title":"Quantized Densely Connected U-Nets for Efficient Landmark Localization","arxiv_id":"1808.02194","date":"2018-08-07","proceeding":"ECCV 2018 9","authors":["Zhiqiang Tang","Xi Peng","Shijie Geng","Lingfei Wu","Shaoting Zhang","Dimitris Metaxas"],"abstract":"In this paper, we propose quantized densely connected U-Nets for efficient\nvisual landmark localization. The idea is that features of the same semantic\nmeanings are globally reused across the stacked U-Nets. This dense connectivity\nlargely improves the information flow, yielding improved localization accuracy.\nHowever, a vanilla dense design would suffer from critical efficiency issue in\nboth training and testing. To solve this problem, we first propose order-K\ndense connectivity to trim off long-distance shortcuts; then, we use a\nmemory-efficient implementation to significantly boost the training efficiency\nand investigate an iterative refinement that may slice the model size in half.\nFinally, to reduce the memory consumption and high precision operations both in\ntraining and testing, we further quantize weights, inputs, and gradients of our\nlocalization network to low bit-width numbers. We validate our approach in two\ntasks: human pose estimation and face alignment. The results show that our\napproach achieves state-of-the-art localization accuracy, but using ~70% fewer\nparameters, ~98% less model size and saving ~75% training memory compared with\nother benchmark localizers. The code is available at\nhttps://github.com/zhiqiangdon/CU-Net.","url_abs":"http://arxiv.org/abs/1808.02194v2","url_pdf":"http://arxiv.org/pdf/1808.02194v2.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":"quantized-densely-connected-u-nets-for","repo_url":"https://github.com/zhiqiangdon/CU-Net","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"face-alignment","task_name":"Face Alignment"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/pose-estimation-on-mpii-human-pose","task":"Pose Estimation","dataset":"MPII Human Pose","model":"DU-Net","rank_in_archive_order":19,"of":46,"metrics":{"PCKh-0.5":"91.2"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.02194","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}