Image-to-text retrieval on MSCOCO 48 (test)
68R@1COSMOS
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| COSMOSPre-training Data=Merged-30M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 68 | 87.8 | 92.5 | |
| COSMOSPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 67.5 | 88.2 | 93.5 | |
| COSMOSPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 64.2 | 85.6 | 91.6 | |
| SigLIPPre-training Data=Merged-30M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 64.1 | 86.2 | 91.9 | |
| LiTPre-training Data=2.5B, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 63.4 | 84.3 | 90.3 | |
| CLIPPre-training Data=Merged-30M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 63.2 | 85.6 | 91.7 | |
| SigLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 62.5 | 85.3 | 91.8 | |
| DreamLIPPre-training Data=Merged-30M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 62.3 | 84.5 | 90.3 | |
| CLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 61 | 85 | 90.9 | |
| MetaCLIPPre-training Data=2.5B, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 59.4 | 80.6 | 87.9 | |
| OpenCLIPPre-training Data=LAION-2B, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 59.3 | 82.5 | 89 | |
| OpenCLIPPre-training Data=DataComp-1B, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 58.2 | 80.8 | 88.5 | |
| DreamLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 57 | 80.9 | 88.4 | |
| OpenCLIPPre-training Data=LAION-400M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 56.5 | 80.4 | 87.3 | |
| CLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 56.2 | 80.4 | 88 | |
| SigLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 55.4 | 79.7 | 87.8 | |
| DreamLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 54 | 78.2 | 86.4 | |
| COSMOSPre-training Data=CC3M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 53.1 | 78.3 | 86.7 | |
| LaSF-CLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 44.3 | 71.3 | 80.2 | |
| SF-CLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 41.8 | 68.3 | 78.4 | |
| MaskCLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 41.4 | 67.9 | 77.5 | |
| DreamLIPPre-training Data=CC3M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 40.8 | 68.5 | 79.4 | |
| CLIPPre-training Data=CC3M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 40.2 | 67 | 77.5 | |
| SigLIPPre-training Data=CC3M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 40.1 | 67.5 | 78.1 | |
| LaCLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 38 | 64.8 | 75 | |
| SLIPPre-training Data=CC12M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 37.6 | 64.9 | 75.5 | |
| MLLM-APre-training Data=CC3M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 35.9 | 62.4 | 73.9 | |
| SLIPPre-training Data=YFCC15M, Evaluation Protocol=Zero-shot, Vision Encoder Architecture=ViT-B/162024.12 | 33.4 | 59.8 | 70.6 |