Dataset Card for WSDMCup2023

Question	Image and Answer
What do you use to hit the ball?
What do people use for cutting?
What do we use to support the immune system and get vitamin C?

Dataset Summary

The WSDMCup2023 Dataset consists of images associated with textual questions. One entry (instance) in our dataset is a question-image pair labeled with the ground truth coordinates of a bounding box containing the visual answer to the given question. The images were obtained from a CC BY-licensed subset of the Microsoft Common Objects in Context dataset, MS COCO. All data labeling was performed on the Toloka crowdsourcing platform.

Our dataset has 45,199 instances split among three subsets: train (38,990 instances), public test (1,705 instances), and private test (4,504 instances). The entire train dataset was available for everyone since the start of the challenge. The public test dataset was available since the evaluation phase of the competition but without any ground truth labels. After the end of the competition, public and private sets were released.

Dataset Citation

Please cite the challenge results or dataset description as follows.

• Ustalov D., Pavlichenko N., Likhobaba D., and Smirnova A. WSDM Cup 2023 Challenge on Visual Question Answering. Proceedings of the 4th Crowd Science Workshop on Collaboration of Humans and Learning Algorithms for Data Labeling. Singapore, 2023, pp. 1–7.

@inproceedings{TolokaWSDMCup2023,
  author    = {Ustalov, Dmitry and Pavlichenko, Nikita and Likhobaba, Daniil and Smirnova, Alisa},
  title     = {{WSDM~Cup 2023 Challenge on Visual Question Answering}},
  year      = {2023},
  booktitle = {Proceedings of the 4th Crowd Science Workshop on Collaboration of Humans and Learning Algorithms for Data Labeling},
  pages     = {1--7},
  address   = {Singapore},
  issn      = {1613-0073},
  url       = {http://ceur-ws.org/Vol-3357/invited1.pdf},
  language  = {english},
}

Supported Tasks and Leaderboards

The Visual Question Answering.

Language

English

Dataset Structure

Data Instances

A data instance contains a URL to the picture, information about the image size - width and height, information about the ground truth bounding box - left top and right bottom dots, and contains the question related to the picture.

{'image': https://toloka-cdn.azureedge.net/wsdmcup2023/000000000013.jpg,
'width': 640,
'height': 427,
'left': 129,
'top': 192,
'right': 155,
'bottom': 212,
'question': What does it use to breath?}

Data Fields

• image: contains URL to the image
• width: value in pixels of image width
• height: value in pixels of image height
• left: the x coordinate in pixels to determine the left-top dot of the bounding box
• top: the y coordinate in pixels to determine the left-top dot of the bounding box
• right: the x coordinate in pixels to determine the right-bottom dot of the bounding box
• bottom: the y coordinate in pixels to determine the right-bottom dot of the bounding box
• question: a question related to the picture

Data Splits

There are four splits in the data: train, train_sample, test_public, and test_private. 'train' split contains the full pull for model training. The 'train-sample' split contains the part of the 'train' split. The 'test_public' split contains public data to test the model. The 'test_private' split contains private data for the final model test.

Source Data

The images were obtained from a CC BY-licensed subset of the Microsoft Common Objects in Context dataset, MS COCO.

Annotations

All data labeling was performed on the Toloka crowdsourcing platform.

Only annotators who self-reported the knowledge of English had access to the annotation task.

Citation Information

• Competition: https://toloka.ai/challenges/wsdm2023
• CodaLab: https://codalab.lisn.upsaclay.fr/competitions/7434
• Dataset: https://doi.org/10.5281/zenodo.7057740