Show, Reward and Tell

这是AAAI2018用GAN和reinforcement learning(RL)做Photo Stream Story Telling的文章。paper链接https://pdfs.semanticscholar.org/977b/eecdf0b5c3487d03738cff501c79770f0858.pdf,暂时还没有找到作何的主页和相关的code,文章题目Show, Reward and Tell: Automatic Generation of Narrative Paragraph from Photo Stream by Adversarial Training
个人瞎扯:看这篇文章的两个原因

  • 这个task算是跨媒体任务中比较有意思的。
  • 这篇paper同时利用了GAN和reinforcement learning(RL)

文章要做的事情(Photo Stream Story Telling):
输入:photo strean(several images)    输出:paragraph
文章show出来的example如下图所示。
example
与state-of-the-art方法和ground-truth对比结果如下所示。
results

method
paper的framework如下所示。
framework

文章中的几个点:
Multi-modal Discriminator(sentence-level): generate relevance sentence of image。判断的内容为图像分别与成对,不成对和生成的sentence做concatenation(discriminate concatenation)。

Language-style Discriminator(paragraph-level): generate human-level story。判断的内容为ground truth stories (gt), random combinations of ground truth sentences (random) and the generated narrative paragraphs (generated)(discriminate paragraph)。

Reward Function: 对relevant sentence和human-level story做reward。