VizWiz – Visual Question Answering

Overview

The VizWiz-VQA (Visual Question Answering) dataset contains 20,523 training images, 8,000 test images, and 4,319 validation images. Each image in the training and validation set has a question about that image and 10 associated answers to the question.

Dataset Metadata

Format License Domain Number of Records Size
JSON
CC BY 4.0 Visual Question Answering 32842 images, 248420 question answer pairs
17.5 GB

Citation

@inproceedings{vizwiz,
author="Danna Gurari and Qing Li and Chi Lin and Yinan Zhao and Anhong Guo and Abigale J. Stangl and Jeffrey P. Bigham",
title="{VizWiz-Priv}: {A} Dataset for Recognizing the Presence and Purpose of Private Visual Information in Images Taken by Blind People",
year=2019,
booktitle={IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}
}