KVQA dataset contains 24602 images harvested from Wikipedia,
associated captions obtained from Wikipedia, named entities
present in the image, their Wikidata ids (Qids), few questions
about each image, corresponding answers (183K QA pairs in all).
We also provide paraphrased questions and type of questions based
on associated challenges. Further, we provide five train/validation
/test splits for each image.

Getting started
---------------------------------------------------------------
Download instructions:
---------------------
>> mkdir datasets
>> cd datasets
>> wget http://dosa.cds.iisc.ac.in/kvqa/KVQAimgs.tar.gz
>> wget http://dosa.cds.iisc.ac.in/kvqa/dataset.json
>> wget http://dosa.cds.iisc.ac.in/kvqa/KVQArefImgs.tar.gz
>> wget http://dosa.cds.iisc.ac.in/kvqa/loadDataset.py


Getting info about dataset:
---------------------------
To load KVQA dataset please run following script.
>> python loadDataset.py

dataset contains image ids as keys. Each key contain
following fields

'imgPath': Image path 
'Qids': Wikidata IDs corresponding to named entities present in image
'NamedEntities': Named entities present in image
'wikiCap': Caption for the image obtained from Wikipedia
'Questions': A list of questions about the image
'ParaQuestions': Corresponding paraphrased questions
'Answers': Corresponding answers
'Type of Question': Type of questions based on associated challenges
'split': Five train/validation/test splits. Number 1,2 and 3 represent
if this image is going to be used as train (1), validation (2) or test(3).

Example:
-----------------------------------------------------------
Example to get different information about one image (img id: 
40392):

data['40392']['imgPath']
'KVQAimgs/40392.jpg'

data['40392']['wikiCap']
'Serena Williams and Venus Williams, Australian Open 2009'

len(data['40392']['Questions']) ## There are total 15 questions about this image
15

data['40392']['Questions'][2] ## Question Number 3
'Who is to the left of Venus Williams?'

data['40392']['ParaQuestions'][2] ## Corresponding paraphrased question
'Who is to the left side of Venus Williams?'

data['40392']['Answers'][2] ## Answer of Question Number 3
'Serena Williams'

data['40392']['Questions'][12] ## Question Number 13
'Who among the people in the image ever married to Alexis Ohanian?'

data['40392']['ParaQuestions'][12] ## Corresponding paraphrased question
'Who among the people in the picture at any point wedded to Alexis Ohanian?'

data['40392']['Answers'][12] ##Answer of Question Number 13
'Person in the right'

data['40392']['split'] ## Five train/test/val splits. This image should
be used as 1=train in first split, 1=train in second split, 1=train in
third split, 1=train in fourth split, 2=validation in fifth split
[1, 1, 1, 1, 2]


------------------------------------------------------------------
If you use this dataset, please cite our AAAI'18 paper.

@InProceedings{shahMYP19,
  author    = "Sanket Shah, Anand Mishra, Naganad Yadati and Partha Pratim Talukdar",
  title     = "KVQA: Knowledge-Aware Visual Question Answering",
  booktitle = "AAAI",
  year      = "2019",
}
-----------------------------------------------------------------

For any queries about the dataset feel free to contact: anandmishra@iisc.ac.in.
Disclaimer: Although dataset is human verified, but can have minor human error. 
Please feel free to report any such error if you find one. We will be happy to 
clean it and reevaluate our baselines.