Image-Caption Alignment on SugarCrepe (test)
96.9Replace Objectxlm-roberta-large-ViT-H-14
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| xlm-roberta-large-ViT-H-14Source=LAION [35]2024.02 | 96.9 | 86 | 72.1 | 63.8 | 72.1 | 93.1 | 86.1 | |
| ViT-bigG-14Source=LAION [35]2024.02 | 96.7 | 88.1 | 74.8 | 62.2 | 74.9 | 92.2 | 84.5 | |
| xlarge: ViT-L-14Source=DataComp [13]2024.02 | 95.5 | 84.5 | 67 | 65 | 66.8 | 91 | 85 | |
| RN50x64Source=OpenAI [33]2024.02 | 94.5 | 83.5 | 70.6 | 61.8 | 66.7 | 83.3 | 74 | |
| CC:DAESource=LLaVA [25]2024.02 | 91 | 82.1 | 82.2 | 73.6 | 78.8 | 77 | 86 | |
| Meanings as TrajectoriesSource=LLaVA [25], Method Citation=[27]2024.02 | 90.4 | 80.6 | 78.8 | 69.9 | 76.6 | 75.7 | 82.8 | |
| Cond. LikelihoodSource=LLaVA [25]2024.02 | 78.8 | 77.7 | 73.3 | 77.6 | 86 | 36.2 | 76.2 |