発表番号P-15
発表タイトル 条件付き画像キャプショニングに向けた敵対的生成ネットワークの検討
筆者氏名・所属 阿部 佑樹(慶應義塾大学 理工学部)
松森 匠哉(慶應義塾大学 理工学部)
妹尾 卓磨(慶應義塾大学 理工学部)
今井 倫太(慶應義塾大学 理工学部)
アブストラクト One of the challenges of image captioning is selectively generating captions using latent variables,where existing approaches have tackled by using a set of known factors and learning with an annotated dataset. On the other hand,in order to leverage potential latent variables on datasets,a mechanism that automatically learns them is required. In this research,we propose a framework based on generative adversarial networks toward conditional image captioning,a task generating captions conditioned with images and latent variables. In experiments,we demonstrated that our proposed model can learn and leverage latent variables on the image classification with several ground truth labels.
論文 PDFファイル