{"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/hierarchical-binary-cnns-for-landmark","title":"Hierarchical binary CNNs for landmark localization with limited resources","arxiv_id":"1808.04803","date":"2018-08-14","proceeding":null,"authors":["Adrian Bulat","Georgios Tzimiropoulos"],"abstract":"Our goal is to design architectures that retain the groundbreaking\nperformance of Convolutional Neural Networks (CNNs) for landmark localization\nand at the same time are lightweight, compact and suitable for applications\nwith limited computational resources. To this end, we make the following\ncontributions: (a) we are the first to study the effect of neural network\nbinarization on localization tasks, namely human pose estimation and face\nalignment. We exhaustively evaluate various design choices, identify\nperformance bottlenecks, and more importantly propose multiple orthogonal ways\nto boost performance. (b) Based on our analysis, we propose a novel\nhierarchical, parallel and multi-scale residual architecture that yields large\nperformance improvement over the standard bottleneck block while having the\nsame number of parameters, thus bridging the gap between the original network\nand its binarized counterpart. (c) We perform a large number of ablation\nstudies that shed light on the properties and the performance of the proposed\nblock. (d) We present results for experiments on the most challenging datasets\nfor human pose estimation and face alignment, reporting in many cases\nstate-of-the-art performance. (e) We further provide additional results for the\nproblem of facial part segmentation. Code can be downloaded from\nhttps://www.adrianbulat.com/binary-cnn-landmark","url_abs":"http://arxiv.org/abs/1808.04803v1","url_pdf":"http://arxiv.org/pdf/1808.04803v1.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":"hierarchical-binary-cnns-for-landmark","repo_url":"https://github.com/1adrianb/binary-networks-pytorch","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"3d-face-alignment","task_name":"3D Face Alignment"},{"task_slug":"binarization","task_name":"Binarization"},{"task_slug":"face-alignment","task_name":"Face Alignment"},{"task_slug":"pose-estimation","task_name":"Pose Estimation"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/3d-face-alignment-on-aflw2000-3d","task":"3D Face Alignment","dataset":"AFLW2000-3D","model":"Ours","rank_in_archive_order":1,"of":1,"metrics":{"NME":"3.26"},"uses_additional_data":false},{"leaderboard":"/sota/face-alignment-on-aflw-lfpa","task":"Face Alignment","dataset":"AFLW-LFPA","model":"Ours","rank_in_archive_order":3,"of":3,"metrics":{"NME":"3.02"},"uses_additional_data":false},{"leaderboard":"/sota/pose-estimation-on-mpii-human-pose","task":"Pose Estimation","dataset":"MPII Human Pose","model":"x","rank_in_archive_order":43,"of":46,"metrics":{"PCKh-0.5":"81.3"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":"https://app.syntology.ai/?focus=1808.04803","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}