Compositional Chain-of-Thought Prompting for Large Multimodal Models

Benchmark Model Rank Results
visual-reasoning-on-winogroundLLaVA-1.5-CCoT#24Text Score: 42.0Image Score: 35.5Group Score: 22.3
visual-reasoning-on-winogroundLLaVA-1.5#33Text Score: 36.0Image Score: 33.3Group Score: 20.1
visual-reasoning-on-winogroundLLaVA-1.5-ZS-CoT#54Text Score: 28.0Image Score: 22.5Group Score: 12.3
visual-reasoning-on-winogroundInstructBLIP-CCoT#65Text Score: 21.0Image Score: 21.3Group Score: 8.3
visual-reasoning-on-winogroundInstructBLIP-ZS-CoT#77Text Score: 9.3Image Score: 16.3Group Score: 4.0
visual-reasoning-on-winogroundInstructBLIP#78Text Score: 7.0Image Score: 11.5Group Score: 3.3