Hierarchical Question-Image Co-Attention for Visual Question Answering
(Submitted on 31 May 2016 (
v1), last revised 19 Jan 2017 (this version, v5))
A number of recent works have proposed attention models for Visual Question Answering (VQA) that generate spatial maps highlighting image regions relevant to answering the question. In this paper, we argue that in addition to modeling "where to look" or visual attention, it is equally important to model "what words to listen to" or question attention. We present a novel co-attention model for VQA that jointly reasons about image and question attention. In addition, our model reasons about the question (and consequently the image via the co-attention mechanism) in a hierarchical fashion via a novel 1-dimensional convolution neural networks (CNN). Our model improves the state-of-the-art on the VQA dataset from 60.3% to 60.5%, and from 61.6% to 63.3% on the COCO-QA dataset. By using ResNet, the performance is further improved to 62.1% for VQA and 65.4% for COCO-QA.
Submission history
From: Jiasen Lu [ view email][v1] Tue, 31 May 2016 22:02:01 GMT (4284kb,D)
[v2] Thu, 2 Jun 2016 01:51:13 GMT (3549kb,D)
[v3] Wed, 26 Oct 2016 02:15:57 GMT (3669kb,D)
[v4] Fri, 13 Jan 2017 16:18:03 GMT (3669kb,D)
[v5] Thu, 19 Jan 2017 05:03:33 GMT (3669kb,D)