Equivariant Similarity for Vision-Language Foundation Models

Benchmark Model Rank Results
visual-reasoning-on-winogroundFIBER (EqSim)#5Text Score: 51.5Image Score: 32.00Group Score: 27.5
visual-reasoning-on-winogroundFIBER (finetuned, Flickr30k)#6Text Score: 51.25Image Score: 26.50Group Score: 23.00
visual-reasoning-on-winogroundFIBER#13Text Score: 46.25Image Score: 25.75Group Score: 22.25
visual-reasoning-on-winogroundMETER (EqSim)#15Text Score: 45.0Image Score: 22.75Group Score: 18.75
visual-reasoning-on-winogroundMETER (finetuned, Flickr30k)#20Text Score: 43.5Image Score: 20.75Group Score: 14.75
visual-reasoning-on-winogroundMETER#26Text Score: 39.25Image Score: 15.75Group Score: 12.00