VQA V1 PyTorch Implementation

This repository contains the code and data necessary to replicate the experiments from the paper VQA: Visual Question Answering. The aim is to develop a model capable of answering questions about images, similar to the system presented in the paper.

Model architecture

Steps to Run

1. Download Data

   python download_data.py

2. Preprocess Images

   python preprocess_image.py

3. Create Vocabulary

   python make_vocabulary.py

4. Prepare VQA Inputs

   python make_vqa_inputs.py
   

This is how the datasets firectory should like after finishing the final preprocessing step

datasets
├── Annotations
├── Images
├── Questions
├── Resized_Images
├── test-dev.npy
├── test.npy
├── train_valid.npy
├── train.npy
├── valid.npy
├── vocab_answers.txt
├── vocab_questions.txt\

Train the Model

```bash
python train.py

Acknowledgements

• COCO Dataset: http://cocodataset.org/
• Paper: VQA: Visual Question Answering