README.md

metadata

license: other
pretty_name: VCR v1.0
task_categories:
  - visual-question-answering
  - image-to-text
tags:
  - image
  - visual-commonsense-reasoning
  - vcr
  - r2c
  - arxiv:1811.10830
configs:
  - config_name: image_examples
    data_files:
      - split: train
        path: image_examples/train-*.parquet
      - split: validation
        path: image_examples/validation-*.parquet
      - split: test
        path: image_examples/test-*.parquet
    default: true
  - config_name: questions
    data_files:
      - split: train
        path: questions/train-*.parquet
      - split: validation
        path: questions/validation-*.parquet
      - split: test
        path: questions/test-*.parquet

VCR v1.0

VCR is the Visual Commonsense Reasoning dataset from "From Recognition to Cognition: Visual Commonsense Reasoning" (CVPR 2019).

This Hugging Face version has two loadable configs:

• image_examples: the default viewer-friendly config, one row per unique image, with grouped annotations.
• questions: one row per original VCR question/answer/rationale example.

The original annotation JSONL files are also included under original_annotations/ for legacy compatibility.

License and terms

Use of VCR is governed by the original VCR/AI2 dataset license and website terms:

• https://visualcommonsense.com/license/
• https://visualcommonsense.com/terms/

Citation

@inproceedings{zellers2019vcr,
  author = {Zellers, Rowan and Bisk, Yonatan and Farhadi, Ali and Choi, Yejin},
  title = {From Recognition to Cognition: Visual Commonsense Reasoning},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year = {2019}
}

Content warning

VCR is annotated from movie images and may contain nudity, violence, offensive content, and biased or problematic depictions.