VQA Binarized
- Download train/val annotations, questions, and images from VQA website (http://www.visualqa.org/) and put them in one folder (e.g.
~/vqa-data
). Unzip the files in the same directory. - Make changes to paths in
prepro.sh
depending on where #1 is stored, and run it (chmod +x prepro.sh; ./prepro.sh
). If you stored it in~/vqa-data
, you can just run it as it is. Note thatprepro.sh
runs three python scripts in order. - To train and test with default settings, run:
python main.py --train True
For descriptions about more options, run:python main.py --help