{"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/recurrent-topic-transition-gan-for-visual","title":"Recurrent Topic-Transition GAN for Visual Paragraph Generation","arxiv_id":"1703.07022","date":"2017-03-21","proceeding":"ICCV 2017 10","authors":["Xiaodan Liang","Zhiting Hu","Hao Zhang","Chuang Gan","Eric P. Xing"],"abstract":"A natural image usually conveys rich semantic content and can be viewed from\ndifferent angles. Existing image description methods are largely restricted by\nsmall sets of biased visual paragraph annotations, and fail to cover rich\nunderlying semantics. In this paper, we investigate a semi-supervised paragraph\ngenerative framework that is able to synthesize diverse and semantically\ncoherent paragraph descriptions by reasoning over local semantic regions and\nexploiting linguistic knowledge. The proposed Recurrent Topic-Transition\nGenerative Adversarial Network (RTT-GAN) builds an adversarial framework\nbetween a structured paragraph generator and multi-level paragraph\ndiscriminators. The paragraph generator generates sentences recurrently by\nincorporating region-based visual and language attention mechanisms at each\nstep. The quality of generated paragraph sentences is assessed by multi-level\nadversarial discriminators from two aspects, namely, plausibility at sentence\nlevel and topic-transition coherence at paragraph level. The joint adversarial\ntraining of RTT-GAN drives the model to generate realistic paragraphs with\nsmooth logical transition between sentence topics. Extensive quantitative\nexperiments on image and video paragraph datasets demonstrate the effectiveness\nof our RTT-GAN in both supervised and semi-supervised settings. Qualitative\nresults on telling diverse stories for an image also verify the\ninterpretability of RTT-GAN.","url_abs":"http://arxiv.org/abs/1703.07022v2","url_pdf":"http://arxiv.org/pdf/1703.07022v2.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":[],"tasks":[{"task_slug":null,"task_name":"Generative Adversarial Network"},{"task_slug":null,"task_name":"Image Description"},{"task_slug":"image-paragraph-captioning","task_name":"Image Paragraph Captioning"},{"task_slug":"sentence","task_name":"Sentence"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/image-paragraph-captioning-on-image-paragraph","task":"Image Paragraph Captioning","dataset":"Image Paragraph Captioning","model":"RTT-GAN (Semi + Fully)","rank_in_archive_order":6,"of":10,"metrics":{"BLEU-4":"9.21","CIDEr":"20.36","METEOR":"18.39"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1703.07022","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}