This repository is built upon the code. Futher introduction will be given shortly.
Almost all flags can be set by yourself at utils/config.py
!
git clone --recursive https://github.com/guoyang9/class-imbalance-VQA.git
* python==3.7.7
* nltk==3.4
* bcolz==1.2.1
* tqdm==4.31.1
* numpy==1.18.4
* pytorch==1.4.0
* tensorboardX==2.1
* torchvision==0.6.0
First of all, make all the data in the right position according to the utils/config.py
!
- Please download the VQA-CP datasets in the original paper.
- The image features can be found at the UpDn repo.
- The pre-trained Glove features can be accessed via GLOVE.
-
process questions and dump dictionary:
python tools/create_dictionary.py
-
process answers and question types:
python tools/compute_softscore.py
-
convert image features to h5:
python tools/detection_features_converter.py
python main.py --loss-fn Plain --name test-VQA --gpu 0
python main.py --loss-fn Plain --name test-VQA --eval-only