Simple Token-Level Confidence Improves Caption Correctness

Benchmark Model Rank Results
visual-reasoning-on-winogroundOFA base (ITM)–Text Score: 26.75Image Score: 10.75Group Score: 6.50
visual-reasoning-on-winogroundOFA base (TLC-A)–Text Score: 24.50Image Score: 23.50Group Score: 13.75
visual-reasoning-on-winogroundOFA large (ITM)–Text Score: 30.75Image Score: 10.25Group Score: 7.25
visual-reasoning-on-winogroundOFA large (TLC-A)–Text Score: 29.25Image Score: 27.00Group Score: 17.50
visual-reasoning-on-winogroundOFA tiny (ITM)–Text Score: 22.75Image Score: 7.75Group Score: 4.50
visual-reasoning-on-winogroundOFA tiny (TLC-A)–Text Score: 16.50Image Score: 15.75Group Score: 6.75