{"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-visual-policy-network-for","title":"Context-Aware Visual Policy Network for Sequence-Level Image Captioning","arxiv_id":"1808.05864","date":"2018-08-16","proceeding":null,"authors":["Daqing Liu","Zheng-Jun Zha","Hanwang Zhang","Yongdong Zhang","Feng Wu"],"abstract":"Many vision-language tasks can be reduced to the problem of sequence\nprediction for natural language output. In particular, recent advances in image\ncaptioning use deep reinforcement learning (RL) to alleviate the \"exposure\nbias\" during training: ground-truth subsequence is exposed in every step\nprediction, which introduces bias in test when only predicted subsequence is\nseen. However, existing RL-based image captioning methods only focus on the\nlanguage policy while not the visual policy (e.g., visual attention), and thus\nfail to capture the visual context that are crucial for compositional reasoning\nsuch as visual relationships (e.g., \"man riding horse\") and comparisons (e.g.,\n\"smaller cat\"). To fill the gap, we propose a Context-Aware Visual Policy\nnetwork (CAVP) for sequence-level image captioning. At every time step, CAVP\nexplicitly accounts for the previous visual attentions as the context, and then\ndecides whether the context is helpful for the current word generation given\nthe current visual attention. Compared against traditional visual attention\nthat only fixes a single image region at every step, CAVP can attend to complex\nvisual compositions over time. The whole image captioning model --- CAVP and\nits subsequent language policy network --- can be efficiently optimized\nend-to-end by using an actor-critic policy gradient method with respect to any\ncaption evaluation metric. We demonstrate the effectiveness of CAVP by\nstate-of-the-art performances on MS-COCO offline split and online server, using\nvarious metrics and sensible visualizations of qualitative visual context. The\ncode is available at https://github.com/daqingliu/CAVP","url_abs":"http://arxiv.org/abs/1808.05864v3","url_pdf":"http://arxiv.org/pdf/1808.05864v3.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-visual-policy-network-for","repo_url":"https://github.com/daqingliu/CAVP","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"deep-reinforcement-learning","task_name":"Deep Reinforcement Learning"},{"task_slug":"image-captioning","task_name":"Image Captioning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1808.05864","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}