Revisiting the Role of Language Priors in Vision-Language Models

Benchmark Model Rank Results
visual-reasoning-on-winogroundBLIP (VisualGPTScore, α-tuned)#32Text Score: 36.5Image Score: 21.5Group Score: 16.8
visual-reasoning-on-winogroundBLIP (ITM)#34Text Score: 35.8Image Score: 15.8Group Score: 13.3
visual-reasoning-on-winogroundBLIP (ITC)#53Text Score: 28.0Image Score: 9.0Group Score: 6.5