Vision-Language Compositional Reasoning on Winoground (test)
61.3Text ScoreJAM (Spread)
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| JAM (Spread)Language Backbone=Llama3.2 (1B), Vision Backbone=ResNet502025.07 | 61.3 | — | — | — | — | |
| JAM (Spread)Language Backbone=Gemma2 (2B), Vision Backbone=DINOv2 (ViT-B; 86M)2025.07 | 58.75 | — | — | — | — | |
| JAM (Spread)Language Backbone=Gemma2 (2B), Vision Backbone=ResNet502025.07 | 58.13 | — | — | — | — | |
| JAM (Spread)Language Backbone=Llama3.2 (1B), Vision Backbone=DINOv2 (ViT-B)2025.07 | 57.5 | — | — | — | — | |
| JAM (Spread)Language Backbone=OLMo2 (7B), Vision Backbone=ResNet502025.07 | 57.5 | — | — | — | — | |
| JAM (Spread)Language Backbone=OLMo2 (7B), Vision Backbone=DINOv2 (ViT-B)2025.07 | 55 | — | — | — | — | |
| Pretrained CLIPVision Backbone=ViT-B/322025.07 | 32.8 | — | — | — | — | |
| CLIPBackbone=ViT-L/142023.03 | — | 0.27 | 0.258 | 0.318 | 0.282 | |
| Diffusion Classifier2023.03 | — | 0.461 | 0.292 | 0.402 | 0.385 | |
| OpenCLIPBackbone=ViT-H/142023.03 | — | 0.39 | 0.266 | 0.335 | 0.33 | |
| Random Chance2023.03 | — | 0.25 | 0.25 | 0.25 | 0.25 |