{"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/learning-deep-visual-object-models-from-noisy","title":"Learning Deep Visual Object Models From Noisy Web Data: How to Make it Work","arxiv_id":"1702.08513","date":"2017-02-28","proceeding":"IEEE Xplore: 2017 12","authors":["Nizar Massouh","Francesca Babiloni","Tatiana Tommasi","Jay Young","Nick Hawes","Barbara Caputo"],"abstract":"Deep networks thrive when trained on large scale data collections. This has\ngiven ImageNet a central role in the development of deep architectures for\nvisual object classification. However, ImageNet was created during a specific\nperiod in time, and as such it is prone to aging, as well as dataset bias\nissues. Moving beyond fixed training datasets will lead to more robust visual\nsystems, especially when deployed on robots in new environments which must\ntrain on the objects they encounter there. To make this possible, it is\nimportant to break free from the need for manual annotators. Recent work has\nbegun to investigate how to use the massive amount of images available on the\nWeb in place of manual image annotations. We contribute to this research thread\nwith two findings: (1) a study correlating a given level of noisily labels to\nthe expected drop in accuracy, for two deep architectures, on two different\ntypes of noise, that clearly identifies GoogLeNet as a suitable architecture\nfor learning from Web data; (2) a recipe for the creation of Web datasets with\nminimal noise and maximum visual variability, based on a visual and natural\nlanguage processing concept expansion strategy. By combining these two results,\nwe obtain a method for learning powerful deep object models automatically from\nthe Web. We confirm the effectiveness of our approach through object\ncategorization experiments using our Web-derived version of ImageNet on a\npopular robot vision benchmark database, and on a lifelong object discovery\ntask on a mobile robot.","url_abs":"http://arxiv.org/abs/1702.08513v1","url_pdf":"http://arxiv.org/pdf/1702.08513v1.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":"learning-deep-visual-object-models-from-noisy","repo_url":"https://github.com/nizarmassouh/WebDB","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"none","reach":null}],"tasks":[{"task_slug":"object","task_name":"Object"},{"task_slug":"object-categorization","task_name":"Object Categorization"},{"task_slug":"object-discovery","task_name":"Object Discovery"}],"methods":[{"method_slug":"1x1-convolution","method_name":"1x1 Convolution"},{"method_slug":"auxiliary-classifier","method_name":"Auxiliary Classifier"},{"method_slug":"average-pooling","method_name":"Average Pooling"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"googlenet","method_name":"GoogLeNet"},{"method_slug":"inception-module","method_name":"Inception Module"},{"method_slug":"local-response-normalization","method_name":"Local Response Normalization"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"softmax","method_name":"Softmax"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}