VQA4Mix

Introduction

This repository contains the implementation and experiments for VQA4Mix, focusing on visual question answering (VQA) in mixed datasets. It includes training scripts, inference scripts, and the necessary steps to reproduce the reported results.

---

Setup Instructions

Prerequisites

• Python >= 3.8
• PyTorch >= 1.9.0

Environment Setup

1. Clone this repository:

   git clone https://github.com/godlikeS97/VQA4Mix.git
   cd VQA4Mix

2. Create and activate a virtual environment:

   python3 -m venv vqa4mix_env
   source vqa4mix_env/bin/activate

---

Data Preparation

Obtaining the Data

1. Download the datasets:

   • Specify the datasets used (e.g., COCO, Art Cap).

2. Unzip the downloaded datasets and place them in the data/ directory:

   mkdir data
   mv <downloaded_dataset> data/

3. Preprocess the data by running the preprocessing jupyter Notebook under each corresponding folder in demo:

---

Key Metrics:

• Accuracy in MCQ

---

Example Results

---

Report Paper

Evaluating the Limitations of Generative Models in Image Captioning

Contact

For questions or issues, please create an issue in this repository or contact the authors.

---