C-VQA: A Compositional Split of the Visual Question Answering (VQA) v1.0 Dataset

TL;DR

This paper introduces C-VQA, a new compositional split of the VQA v1.0 dataset, to evaluate models' ability to generalize to unseen concept combinations, revealing significant performance drops of existing models.

Contribution

It proposes a novel compositional split for VQA, enabling assessment of models' compositional understanding beyond superficial correlations.

Findings

Existing models perform significantly worse on C-VQA compared to original VQA.

C-VQA highlights the lack of compositional generalization in current VQA models.

The dataset facilitates future research on compositional reasoning in VQA.

Abstract

Visual Question Answering (VQA) has received a lot of attention over the past couple of years. A number of deep learning models have been proposed for this task. However, it has been shown that these models are heavily driven by superficial correlations in the training data and lack compositionality -- the ability to answer questions about unseen compositions of seen concepts. This compositionality is desirable and central to intelligence. In this paper, we propose a new setting for Visual Question Answering where the test question-answer pairs are compositionally novel compared to training question-answer pairs. To facilitate developing models under this setting, we present a new compositional split of the VQA v1.0 dataset, which we call Compositional VQA (C-VQA). We analyze the distribution of questions and answers in the C-VQA splits. Finally, we evaluate several existing VQA models under this new setting and show that the performances of these models degrade by a significant amount compared to the original VQA setting.