Skip to content
Structured Attentions for Visual Question Answering
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Structured Attentions for Visual Question Answering

The repository contains the majority of the code to reproduce the experimental results of the paper Structured Attentions for Visual Question Answering on the VQA-1.0 and VQA-2.0 dataset. Currently only the accelerated version of Mean Field is provided, which is used in the VQA 2.0 challenge.


The framework of the proposed network.


To reproduce the experimental results,

  • Clone and compile mxnet, with mxnet@c9e252, cub@89de7ab, dmlc-core@3dfbc6, nnvm@d3558d, ps-lite@acdb69, mshadow@8eb1e0. There has been some modification on optimizers (and others) in later versions of mxnet, and code in this repository has not been adapted yet.

  • ResNet-152 feature of MS COCO images: extracted with MCB's preprocess code.

  • Our training question and answer data for VQA2.0: Baidu Pan.

Training from scratch

Set the arguments and run

Pretrained models

The best single model accuracy on test-dev of VQA-1.0 and VQA-2.0 with skip-thought vector initialization and Visual Genome training data are 67.19 and 64.78 respectively. Here is the model on VQA-2.0.


If you found this repository helpful, you could cite

  title={Structured Attentions for Visual Question Answering},
  author={Chen, Zhu and Yanpeng, Zhao and Shuaiyi, Huang and Kewei, Tu and Yi, Ma},
  journal={IEEE International Conference on Computer Vision (ICCV)},


This code is distributed under MIT LICENSE.

You can’t perform that action at this time.