Simple Baseline for Visual Question Answering

We descrive a very simple bag-of-words baseline for visual question answering. The description of the baseline is in the arXiv paper http://arxiv.org/pdf/1512.02167.pdf. The code is developed by Bolei Zhou and Yuandong Tian.

Demo is available at http://visualqa.csail.mit.edu/

To train the model using the code, the following data of the VQA dataset are needed:

• The pre-processed data of text is at http://visualqa.csail.mit.edu/data_vqa_txt.zip
• The googlenet feature of all the COCO images is at http://visualqa.csail.mit.edu/data_vqa_feat.zip

Contact Bolei Zhou ([email protected]) if you have any questions.

Please cite our arXiv note if you use our code:

B. Zhou, Y. Tian, S. Suhkbaatar, A. Szlam, R. Fergus. Simple Baseline for Visual Question Answering. arXiv:1512.02167