A Gaze-grounded Visual Question Answering Dataset for Clarifying Ambiguous Japanese Questions

TL;DR

This paper introduces GazeVQA, a new dataset and method that use gaze information to clarify ambiguous Japanese visual questions, improving VQA accuracy and addressing language-specific ambiguities.

Contribution

The study presents a novel GazeVQA dataset and a gaze target estimation method to enhance visual question answering for ambiguous Japanese questions.

Findings

GazeVQA improves VQA accuracy in some cases.

Gaze information helps clarify ambiguous questions.

Identifies challenges in GazeVQA tasks.

Abstract

Situated conversations, which refer to visual information as visual question answering (VQA), often contain ambiguities caused by reliance on directive information. This problem is exacerbated because some languages, such as Japanese, often omit subjective or objective terms. Such ambiguities in questions are often clarified by the contexts in conversational situations, such as joint attention with a user or user gaze information. In this study, we propose the Gaze-grounded VQA dataset (GazeVQA) that clarifies ambiguous questions using gaze information by focusing on a clarification process complemented by gaze information. We also propose a method that utilizes gaze target estimation results to improve the accuracy of GazeVQA tasks. Our experimental results showed that the proposed method improved the performance in some cases of a VQA system on GazeVQA and identified some typical problems of GazeVQA tasks that need to be improved.