Prepare Data

Download datasets as you need, and organize them as following:

code_root/
└── data/
   └── coco/
       ├── train2014/
       ├── val2014/
       ├── test2015/
       ├── train2017/
       ├── val2017/
       ├── test2017/
       ├── annotations/
       ├── aokvqa/
           ├── commonsense
               └── expansions
       ├── okvqa/
           ├── commonsense
               └── expansions
       ├── vqa/
       └── vgbua_res101_precomputed/
           ├── trainval2014_resnet101_faster_rcnn_genome
           └── test2015_resnet101_faster_rcnn_genome
       └── sbert/
           ├── aokvqa
           └── okvqa 
       

Data

VQA & RefCOCO+

Common

• Download and unzip COCO 2014 images & annotations from here.
• For A-OKVQA, we use COCO 2017 images, which you can also download at the above link.

VQA

• Download and unzip annotations from here (including "VQA Annotations" and "VQA Input Questions"), place all these files directly under ./data/coco/vqa.

• Download and unzip following precomputed boxes & features into ./data/coco/vgbua_res101_precomputed.

  • train2014 + val2014: GoogleDrive / BaiduPan
  • test2015: GoogleDrive / BaiduPan

• Download answer vocabulary from GoogleDrive / BaiduPan, place it under the folder ./data/coco/vqa/.

OK-VQA

• Download the training and testing files from here and save them in the ./data/coco/okvqa folder

A-OKVQA

• Download the training and testing files from here and save them in the ./data/coco/aokvqa folder