{"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/towards-open-set-deep-networks","title":"Towards Open Set Deep Networks","arxiv_id":"1511.06233","date":"2015-11-19","proceeding":"CVPR 2016 6","authors":["Abhijit Bendale","Terrance Boult"],"abstract":"Deep networks have produced significant gains for various visual recognition\nproblems, leading to high impact academic and commercial applications. Recent\nwork in deep networks highlighted that it is easy to generate images that\nhumans would never classify as a particular object class, yet networks classify\nsuch images high confidence as that given class - deep network are easily\nfooled with images humans do not consider meaningful. The closed set nature of\ndeep networks forces them to choose from one of the known classes leading to\nsuch artifacts. Recognition in the real world is open set, i.e. the recognition\nsystem should reject unknown/unseen classes at test time. We present a\nmethodology to adapt deep networks for open set recognition, by introducing a\nnew model layer, OpenMax, which estimates the probability of an input being\nfrom an unknown class. A key element of estimating the unknown probability is\nadapting Meta-Recognition concepts to the activation patterns in the\npenultimate layer of the network. OpenMax allows rejection of \"fooling\" and\nunrelated open set images presented to the system; OpenMax greatly reduces the\nnumber of obvious errors made by a deep network. We prove that the OpenMax\nconcept provides bounded open space risk, thereby formally providing an open\nset recognition solution. We evaluate the resulting open set deep networks\nusing pre-trained networks from the Caffe Model-zoo on ImageNet 2012 validation\ndata, and thousands of fooling and open set images. The proposed OpenMax model\nsignificantly outperforms open set recognition accuracy of basic deep networks\nas well as deep networks with thresholding of SoftMax probabilities.","url_abs":"http://arxiv.org/abs/1511.06233v1","url_pdf":"http://arxiv.org/pdf/1511.06233v1.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":"towards-open-set-deep-networks","repo_url":"https://github.com/Andrewwango/open-set-classif","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"towards-open-set-deep-networks","repo_url":"https://github.com/Andrewwango/open-set-resnet","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}},{"paper_slug":"towards-open-set-deep-networks","repo_url":"https://github.com/geoai-lab/MapMetadataEnrichment","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"tf","reach":{"status":"ok"}},{"paper_slug":"towards-open-set-deep-networks","repo_url":"https://github.com/aadeshnpn/OSDN","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":{"status":"ok","spdx":"LGPL-3.0"}}],"tasks":[{"task_slug":"open-set-learning","task_name":"Open Set Learning"}],"methods":[{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1511.06233","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}