Vision Skills Needed to Answer Visual Questions

TL;DR

This paper analyzes essential vision skills for visual question answering, compares human and AI performance, and introduces a new task to predict required skills, highlighting gaps between user needs and AI focus.

Contribution

It identifies key vision skills needed for VQA, evaluates their difficulty for humans and AI, and proposes a novel skill prediction task to improve understanding and development.

Findings

Object, text, color recognition, and counting are crucial for VQA.

Humans outperform AI in certain skills, revealing gaps.

Proposed skill prediction task offers new insights for VQA development.

Abstract

The task of answering questions about images has garnered attention as a practical service for assisting populations with visual impairments as well as a visual Turing test for the artificial intelligence community. Our first aim is to identify the common vision skills needed for both scenarios. To do so, we analyze the need for four vision skills---object recognition, text recognition, color recognition, and counting---on over 27,000 visual questions from two datasets representing both scenarios. We next quantify the difficulty of these skills for both humans and computers on both datasets. Finally, we propose a novel task of predicting what vision skills are needed to answer a question about an image. Our results reveal (mis)matches between aims of real users of such services and the focus of the AI community. We conclude with a discussion about future directions for addressing the visual question answering task.