VCR v1.0

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.… See the full description on the dataset page:

Source: Hugging Face Hub (Rowan/vcr). Metadata imported from the dataset’s Hub tags.