Visual question and answer

This model is based off https://github.com/Cyanogenoid/pytorch-vqa Before running the code please download the relevant train, validation and test data from https://vizwiz.org/tasks-and-datasets/vqa/ Each section below tells you which scripts to run depending on whether the results for the Resnet-18 or CNN are desired

We run the scripts in the following order to achieve the best results

image_preprocessing_CNN.py

This extracts features from the image

preprocess-QA.py

This extracts features from the questions asked

train_models_cnn.py

This trains the model for 50 epochs with L2 regularization 0.0005

view-logs.py logs/<log name>

This prints out the results

We run the scripts in the following order to achieve the best results

image_preprocessing_resnet18.py

This extracts features from the image

preprocess-QA.py

This extracts features from the questions asked

train_models.py

This trains the model for 50 epochs with L2 regularization 0.001

view-logs.py logs/<log name>

This prints out the results