{"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/learn-to-pay-attention","title":"Learn To Pay Attention","arxiv_id":"1804.02391","date":"2018-04-06","proceeding":"ICLR 2018 1","authors":["Saumya Jetley","Nicholas A. Lord","Namhoon Lee","Philip H. S. Torr"],"abstract":"We propose an end-to-end-trainable attention module for convolutional neural\nnetwork (CNN) architectures built for image classification. The module takes as\ninput the 2D feature vector maps which form the intermediate representations of\nthe input image at different stages in the CNN pipeline, and outputs a 2D\nmatrix of scores for each map. Standard CNN architectures are modified through\nthe incorporation of this module, and trained under the constraint that a\nconvex combination of the intermediate 2D feature vectors, as parameterised by\nthe score matrices, must \\textit{alone} be used for classification.\nIncentivised to amplify the relevant and suppress the irrelevant or misleading,\nthe scores thus assume the role of attention values. Our experimental\nobservations provide clear evidence to this effect: the learned attention maps\nneatly highlight the regions of interest while suppressing background clutter.\nConsequently, the proposed function is able to bootstrap standard CNN\narchitectures for the task of image classification, demonstrating superior\ngeneralisation over 6 unseen benchmark datasets. When binarised, our attention\nmaps outperform other CNN-based attention maps, traditional saliency maps, and\ntop object proposals for weakly supervised segmentation as demonstrated on the\nObject Discovery dataset. We also demonstrate improved robustness against the\nfast gradient sign method of adversarial attack.","url_abs":"http://arxiv.org/abs/1804.02391v2","url_pdf":"http://arxiv.org/pdf/1804.02391v2.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":"learn-to-pay-attention","repo_url":"https://github.com/caoquanjie/LearnToPayAttention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"learn-to-pay-attention","repo_url":"https://github.com/mb010/AstroAttention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null},{"paper_slug":"learn-to-pay-attention","repo_url":"https://github.com/wangyongjie-ntu/Learn-to-pay-attention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"learn-to-pay-attention","repo_url":"https://github.com/SaoYan/LearnToPayAttention","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"GPL-3.0"}}],"tasks":[{"task_slug":"adversarial-attack","task_name":"Adversarial Attack"},{"task_slug":"classification","task_name":"General Classification"},{"task_slug":"image-classification","task_name":"Image Classification"},{"task_slug":"object-discovery","task_name":"Object Discovery"},{"task_slug":"weakly-supervised-segmentation","task_name":"Weakly supervised segmentation"},{"task_slug":"image-classification","task_name":"image-classification"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1804.02391","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}