{"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/grounding-of-textual-phrases-in-images-by","title":"Grounding of Textual Phrases in Images by Reconstruction","arxiv_id":"1511.03745","date":"2015-11-12","proceeding":null,"authors":["Anna Rohrbach","Marcus Rohrbach","Ronghang Hu","Trevor Darrell","Bernt Schiele"],"abstract":"Grounding (i.e. localizing) arbitrary, free-form textual phrases in visual\ncontent is a challenging problem with many applications for human-computer\ninteraction and image-text reference resolution. Few datasets provide the\nground truth spatial localization of phrases, thus it is desirable to learn\nfrom data with no or little grounding supervision. We propose a novel approach\nwhich learns grounding by reconstructing a given phrase using an attention\nmechanism, which can be either latent or optimized directly. During training\nour approach encodes the phrase using a recurrent network language model and\nthen learns to attend to the relevant image region in order to reconstruct the\ninput phrase. At test time, the correct attention, i.e., the grounding, is\nevaluated. If grounding supervision is available it can be directly applied via\na loss over the attention mechanism. We demonstrate the effectiveness of our\napproach on the Flickr 30k Entities and ReferItGame datasets with different\nlevels of supervision, ranging from no supervision over partial supervision to\nfull supervision. Our supervised variant improves by a large margin over the\nstate-of-the-art on both datasets.","url_abs":"http://arxiv.org/abs/1511.03745v4","url_pdf":"http://arxiv.org/pdf/1511.03745v4.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":"grounding-of-textual-phrases-in-images-by","repo_url":"https://github.com/Seth-Park/MultimodalExplanations","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":null},{"paper_slug":"grounding-of-textual-phrases-in-images-by","repo_url":"https://github.com/akirafukui/vqa-mcb","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":{"status":"ok","spdx":"BSD-2-Clause"}},{"paper_slug":"grounding-of-textual-phrases-in-images-by","repo_url":"https://github.com/divelab/vqa-text","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"caffe2","reach":{"status":"ok"}}],"tasks":[{"task_slug":"language-modeling","task_name":"Language Modeling"},{"task_slug":"language-modelling","task_name":"Language Modelling"},{"task_slug":"natural-language-visual-grounding","task_name":"Natural Language Visual Grounding"},{"task_slug":"phrase-grounding","task_name":"Phrase Grounding"},{"task_slug":"visual-grounding","task_name":"Visual Grounding"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/phrase-grounding-on-flickr30k-entities-test","task":"Phrase Grounding","dataset":"Flickr30k Entities Test","model":"GroundeR 100.0% annot.","rank_in_archive_order":13,"of":18,"metrics":{"R@1":"48.38"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1511.03745","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}