gqa_pytorch

Basic model for visual question answering
Data: https://cs.stanford.edu/people/dorarad/gqa/index.html
Major part of preprocess.py and dataset.py is from https://github.com/ronilp/mac-network-pytorch-gqa
Attention model for VQA: https://arxiv.org/abs/1707.07998 Simplify some functions to accelerate the process.

notes:

• train_all.py is training file for basic CNN+LSTM model
• train_att.py is training file for attention model
• train_double_att.py is training file for double attention model
• model.py is the model file for basic model
• attention.py is the model file for attention model
• attention2.py is the model file for double attention model

TODO:

1. TEST.py (waiting for the official test set releasing)
2. Attention. (done)
3. Add a channel for Graph.
4. Try the idea of Bi-Directional Attention Flow. (done)