论文阅读：Learning Visual Question Answering by Bootstrapping Hard Attention

Learning Visual Question Answering by Bootstrapping Hard Attention

Google DeepMind  ECCV-2018

  2018-08-05 19:24:44

 

Paper：https://arxiv.org/abs/1808.00300 

 

Introduction：

本文尝试仅仅用 hard attention 的方法来抠出最有用的 feature，进行 VQA 任务的学习。

Soft Attention： 　　

　　Existing attention models [7,8,9,10] are predominantly based on soft attention, in which all information is adaptively re-weighted before being aggregated. This can improve accuracy by isolating important information and avoiding interference from unimportant information. 

Hard Attention：

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

--