Language Compositionality on SugarCrepe (test)
100Replace: Object (R@1)Human
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| HumanParams (B)=-2024.12 | 100 | 99 | 97 | 99 | 100 | 99 | 99 | — | — | — | |
| VladVA (Ours)Params (B)=7.06, Backbone=LLaVA-1.5-7B2024.12 | 98.1 | 92.1 | 86.8 | 79 | 82.9 | 95.2 | 95.8 | — | — | — | |
| BLIP2 (ViT-L)Params (B)=1.17, Backbone=ViT-L2024.12 | 97.6 | 81.7 | 77.8 | 62.1 | 65.5 | 92.4 | 87.4 | — | — | — | |
| EVA-CLIP (18B)Params (B)=18.32024.12 | 97.5 | 88.8 | 76.1 | 65.3 | 76 | 92.4 | 85 | — | — | — | |
| VLM2Vec (Mistral-7B)Params (B)=7.30, Backbone=Mistral-7B2024.12 | 97.2 | 89 | 81.7 | 62.9 | 72.5 | 94.7 | 88.6 | — | — | — | |
| EVA-02-CLIP (ViT-E/14+)Params (B)=5.04, Backbone=ViT-E/14+2024.12 | 97.1 | 88.5 | 74.2 | 67.3 | 74.1 | 91.8 | 83.9 | — | — | — | |
| E5-V (LLaVA-Next-8B)Params (B)=8.36, Backbone=LLaVA-Next-8B2024.12 | 96.7 | 89.5 | 85.3 | 75 | 70.1 | 89.2 | 83.5 | — | — | — | |
| OpenCLIP (ViT-bigG/14)Params (B)=2.54, Backbone=ViT-bigG/142024.12 | 96.6 | 87.9 | 74.9 | 62.5 | 75.2 | 92.2 | 84.5 | — | — | — | |
| BLIP (ViT-L)Params (B)=0.23, Backbone=ViT-L2024.12 | 96.5 | 81.7 | 69.1 | 66.6 | 76.8 | 92 | 85.1 | — | — | — | |
| EVA-CLIP (8B)Params (B)=8.222024.12 | 96.4 | 86.6 | 74.8 | 66.1 | 74.6 | 91.3 | 82 | — | — | — | |
| OpenCLIP (ViT-G/14)Params (B)=1.37, Backbone=ViT-G/142024.12 | 95.8 | 85 | 72.4 | 63 | 71.2 | 91.5 | 82.1 | — | — | — | |
| E5-V (LLaVA-1.5-7B)Params (B)=7.06, Backbone=LLaVA-1.5-7B2024.12 | 95.8 | 86.6 | 81.6 | 62.9 | 64 | 93.5 | 88 | — | — | — | |
| CLIP (ViT-L)Params (B)=0.43, Backbone=ViT-L2024.12 | 94.1 | 79.2 | 65.2 | 60.2 | 62.3 | 78.3 | 71.5 | — | — | — | |
| NegCLIPParams (B)=0.152024.12 | 92.7 | 85.9 | 76.5 | 75.2 | 75.4 | 88.8 | 82.8 | — | — | — | |
| CLIP-PGS0.3Relative Training Time=0.6×2025.03 | 88.1 | 76 | 67.9 | 64.1 | 66.5 | 74.2 | 69.9 | 75.5 | 70.8 | 67.9 | |
| LLaVA-1.5-7BParams (B)=7.062024.12 | 88 | 81.6 | 76.1 | 60.9 | 58.8 | 67 | 62.4 | — | — | — | |
| E-CLIPRelative Training Time=0.6x2025.03 | 86.9 | 73.5 | 60.2 | 59.4 | 63.4 | 73.3 | 66.8 | 73.2 | 68.4 | 60.2 | |
| A-CLIPRelative Training Time=1.1x2025.03 | 86.6 | 75.5 | 63.2 | 52.4 | 63.1 | 71.6 | 66.8 | 71.6 | 68.4 | 63.2 | |
| CLIP-PGS0.5Relative Training Time=0.5×2025.03 | 86 | 77 | 64.6 | 63.3 | 65.5 | 77.3 | 69.8 | 75.5 | 70.8 | 64.6 | |
| CLIPRelative Training Time=1.0×2025.03 | 85.8 | 79.2 | 64.5 | 61.8 | 58.7 | 74.2 | 68.4 | 73.7 | 68.8 | 64.5 | |
| FLIPRelative Training Time=0.5×2025.03 | 84.1 | 75.9 | 66 | 60.2 | 61.6 | 71.7 | 63.2 | 72 | 66.9 | 66 |