VCR_en_easy

Visual Commonsense Reasoning (VCR) benchmark that tests higher-order cognition and commonsense reasoning beyond simple object recognition. Models must answer challenging questions about images and provide rationales justifying their answers. The benchmark measures the ability to infer people's actions, goals, and mental states from visual context.

id vcr-en-easy · max 1 · 1 models reported