[Paper Reading] Show, Attend and Tell: Neural Image Caption Generation with Visual Attention

Posted zlian2016

tags:

篇首语:本文由小常识网(cha138.com)小编为大家整理,主要介绍了[Paper Reading] Show, Attend and Tell: Neural Image Caption Generation with Visual Attention相关的知识,希望对你有一定的参考价值。

论文链接:https://arxiv.org/pdf/1502.03044.pdf

代码链接:https://github.com/kelvinxu/arctic-captions & https://github.com/yunjey/show-attend-and-tell & https://github.com/jazzsaxmafia/show_attend_and_tell.tensorflow


主要贡献

在这篇文章中,作者将“注意力机制(Attention Mechanism)”引入了神经机器翻译(Neural Image Captioning)领域,提出了两种不同的注意力模型:‘Soft’ Deterministic Attention Mechanism & ‘Hard’ Stochastic Attention Mechanism。下图展示了Show, Attend and Tell模型的整体框架。

技术图片

  • Stochastic ‘Hard‘ Attention
  • Deterministic ‘Soft‘ Attention

实验细节

 

以上是关于[Paper Reading] Show, Attend and Tell: Neural Image Caption Generation with Visual Attention的主要内容,如果未能解决你的问题,请参考以下文章

Paper Reading 4:Massively Parallel Methods for Deep Reinforcement Learning

Paper Reading - 综述系列 - 计算机视觉领域中目标检测任务常见问题与解决方案

Paper Reading

Testing & Paper reading——Sketchvisor

CVPR 2016 paper reading

Paper Reading_Computer Architecture