TallyQA: Answering Complex Counting Questions
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


TallyQA Paper link

Please, download the dataset using this link. The zipped file contains json files for train and test respectively. Each entries in the dataset contains the following fields:

 {'answer': 4,
  'data_source': 'imported_genome',
  'image': 'VG_100K_2/2410408.jpg',
  'image_id': 92410408,
  'issimple': False,
  'question': 'How many headlights does the black bus have?',
  'question_id': 30095774}

Attributes such as answer, image-id, image ,question suggest the straighforward meanings. The boolean flag field ( issimple) denotes whether the question is simple (True) or complex ( False) . The field 'data_source' shows the source of the question-answer pair. Questions collected from AMT annotators have data_source = amt and likewise for other sources. All the images used in the dataset are derived from COCO and Visual Genome. The imported QA pairs have been derived from TDIUC and VQA datasets.

Please cite the work using the following Bibtex:

  title={TallyQA: Answering Complex Counting Questions},
  author={Acharya, Manoj and Kafle, Kushal and Kanan, Christopher},