{"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/deep-image-representations-using-caption","title":"Deep image representations using caption generators","arxiv_id":"1705.09142","date":"2017-05-25","proceeding":null,"authors":["Konda Reddy Mopuri","Vishal B. Athreya","R. Venkatesh Babu"],"abstract":"Deep learning exploits large volumes of labeled data to learn powerful\nmodels. When the target dataset is small, it is a common practice to perform\ntransfer learning using pre-trained models to learn new task specific\nrepresentations. However, pre-trained CNNs for image recognition are provided\nwith limited information about the image during training, which is label alone.\nTasks such as scene retrieval suffer from features learned from this weak\nsupervision and require stronger supervision to better understand the contents\nof the image. In this paper, we exploit the features learned from caption\ngenerating models to learn novel task specific image representations. In\nparticular, we consider the state-of-the art captioning system Show and\nTell~\\cite{SnT-pami-2016} and the dense region description model\nDenseCap~\\cite{densecap-cvpr-2016}. We demonstrate that, owing to richer\nsupervision provided during the process of training, the features learned by\nthe captioning system perform better than those of CNNs. Further, we train a\nsiamese network with a modified pair-wise loss to fuse the features learned\nby~\\cite{SnT-pami-2016} and~\\cite{densecap-cvpr-2016} and learn image\nrepresentations suitable for retrieval. Experiments show that the proposed\nfusion exploits the complementary nature of the individual features and yields\nstate-of-the art retrieval results on benchmark datasets.","url_abs":"http://arxiv.org/abs/1705.09142v1","url_pdf":"http://arxiv.org/pdf/1705.09142v1.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":"deep-image-representations-using-caption","repo_url":"https://github.com/mopurikreddy/strong-supervision","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"none","reach":null}],"tasks":[{"task_slug":"retrieval","task_name":"Retrieval"},{"task_slug":"transfer-learning","task_name":"Transfer Learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}