Divide and Conquer: Answering Questions with Object Factorization and Compositional Reasoning

This repository implements the PrOtotypical nEural Module network. It contains four key components, stored in the following directories:

• vqa_exp: zero-shot VQA experiments on the VQA v2 and Novel-VQA datasets
• gqa_exp: zero-shot VQA experiments on the GQA dataset,
• vqa_cp_exp: Out-of-distribution VQA experiments on VQA-CP dataset,
• proto_learning: the code for our prototype learning method with object factorization, and
• data: code for data preprocessing

Please refer to the README in each directory for details.

Disclaimer

We adopt the official implementation of the XNM as the backbone model for prototypical reasoning. We use the bottom-up features provided in the following repos: for VQA and for GQA. Please refer to these links for further README information.

Reference

If you use our code or data, please cite our paper:

@InProceedings{poem,
author = {Chen, Shi and Zhao, Qi},
title = {Divide and Conquer: Answering Questions with Object Factorization and Compositional Reasoning},
booktitle = {CVPR},
year = {2023}
}