ResearchBenchmarksVisual Commonsense Reasoning on VCR QR -> A (val)Follow94.7AccuracyModCR83.7886.61589.4592.285May 8, 2023Evaluation ResultsMethodMethodLinksAccuracyModCRTypes=fine-tune VLMsTypes=fine-tune VLMs2023.0594.7ModCRTypes=frozen VLMsTypes=frozen VLMs2023.0594.5ROBERTa-L2023.0592.7CALEC2023.0590.8OFA-L2023.0590.3Oscar-B2023.0587.3MVPTR2023.0584.2