VizWiz Dataset

VizWiz dataset was collected using the VizWiz application, which was released in May 2011.
  • 11,045 users asked 72,205 visual questions, 48,169 of which were asked by users who agreed to allow their questions to be anonymized and shared. We carefully checked and removed those containing personally identifying information or indecent content, resulting in 31,073 remaining visual questions.
  • Each visual question includes an image, a transcription of the question, and 10 answers crowdsourced from Amazon Mechanical Turk workers. Visual questions originated from blind people who each took a picture using a mobile phone and recorded a spoken question about it.

VizWiz v1.0 dataset download:
  • 20,000 training images
  • 20,000 training questions
  • 200,000 training answers
  • 3,173 validation images
  • 3,173 validation questions
  • 31,730 validation answers
  • 8,000 test images
  • 8,000 test questions


To appear late February 2018

Copyright © 2015