{"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/context-aware-captions-from-context-agnostic","title":"Context-aware Captions from Context-agnostic Supervision","arxiv_id":"1701.02870","date":"2017-01-11","proceeding":"CVPR 2017 7","authors":["Ramakrishna Vedantam","Samy Bengio","Kevin Murphy","Devi Parikh","Gal Chechik"],"abstract":"We introduce an inference technique to produce discriminative context-aware\nimage captions (captions that describe differences between images or visual\nconcepts) using only generic context-agnostic training data (captions that\ndescribe a concept or an image in isolation). For example, given images and\ncaptions of \"siamese cat\" and \"tiger cat\", we generate language that describes\nthe \"siamese cat\" in a way that distinguishes it from \"tiger cat\". Our key\nnovelty is that we show how to do joint inference over a language model that is\ncontext-agnostic and a listener which distinguishes closely-related concepts.\nWe first apply our technique to a justification task, namely to describe why an\nimage contains a particular fine-grained category as opposed to another\nclosely-related category of the CUB-200-2011 dataset. We then study\ndiscriminative image captioning to generate language that uniquely refers to\none of two semantically-similar images in the COCO dataset. Evaluations with\ndiscriminative ground truth for justification and human studies for\ndiscriminative image captioning reveal that our approach outperforms baseline\ngenerative and speaker-listener approaches for discrimination.","url_abs":"http://arxiv.org/abs/1701.02870v3","url_pdf":"http://arxiv.org/pdf/1701.02870v3.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":"context-aware-captions-from-context-agnostic","repo_url":"https://github.com/ruotianluo/DiscCaptioning","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok"}}],"tasks":[{"task_slug":"image-captioning","task_name":"Image Captioning"},{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1701.02870","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}