What AI says about VCR

Visual Commonsense Reasoning (VCR) is a task and large-scale dataset for cognition-level visual understanding, requiring models to answer challenging visual questions and provide rationales explaining their answers. The dataset includes 290k multiple-choice questions with correct answers and rationales, built on 110k images and 80 object categories from COCO.