Tips and Tricks for Visual Question Answering: Learnings from the 2017 Challenge

CVPR2018一篇关于Visual Question Answering Tricks的文章,作者是2017 VQA Challenge冠军团队成员之一,paper连接https://arxiv.org/abs/1708.02711,作者的homepage https://www.damienteney.info/adventures
文章要做的事情:
visual question answer

method
文章的framework如下所示。
Tips and Tricks for Visual Question Answering: Learnings from the 2017 Challenge
主要思路是用bottom-up attention方式得到很多的proposal,然后在用Top-down attention学习这些proposal的权重。