Datasets:

nv-njb
/

EgoTaskQA-MCQ

license: other
license_name: egotaskqa-research-only
license_link: https://sites.google.com/view/egotaskqa
task_categories:
  - visual-question-answering
  - multiple-choice
language:
  - en
size_categories:
  - 10K<n<100K
pretty_name: EgoTaskQA-MCQ
configs:
  - config_name: default
    data_files:
      - split: test
        path: test.json

EgoTaskQA-MCQ

Multiple-choice variant of the EgoTaskQA benchmark (Jia et al., NeurIPS 2022). Each item pairs an egocentric video clip with a 5-way MCQ (A–E) over object states, actions, and goals.

Format

test.json is a list of records:

{
  "video_path": "27k-22-11-2|P1|4546|4986.mp4",
  "q": "What is the status of sandwich before ...",
  "option": {"A": "on top of plate", "B": "...", "C": "...", "D": "...", "E": "..."},
  "a": "A"
}

a is the letter of the correct option.

Videos (not included)

This repository hosts annotations only — the 2,337 egocentric video clips referenced by video_path are not redistributed here, since the original EgoTaskQA dataset is gated under a research-only license.

To obtain the videos:

Visit https://sites.google.com/view/egotaskqa and submit the license-agreement form linked from the Download page.
Once granted access, download the qa_videos/ directory.
Either place the files under ~/.cache/lmms_eval/egotaskqa/videos/, or point the EGOTASKQA_VIDEO_DIR environment variable at the directory.

Filenames in qa_videos/ match the video_path field in this dataset (<scene>-<...>|<participant>|<start>|<end>.mp4); no renaming is required.

License

The MCQ annotations in this repo are released for research use only. Upstream video content remains subject to the EgoTaskQA license; users must obtain it through the official channel above.

Citation

@inproceedings{jia2022egotaskqa,
  title={EgoTaskQA: Understanding Human Tasks in Egocentric Videos},
  author={Jia, Baoxiong and Lei, Ting and Zhu, Song-Chun and Huang, Siyuan},
  booktitle={Thirty-sixth Conference on Neural Information Processing Systems Datasets and Benchmarks Track},
  year={2022}
}