Image Classification on Zero-Shot Evaluation Suite (11 Datasets) (test)
90.5Food-101 Top-1 AccOpenCLIP
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OpenCLIPPre-training Data=DataComp-1B, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 90.5 | 96.3 | 82.2 | 70 | 88.9 | 29.8 | 57.7 | 92.8 | 91.8 | 75.7 | 73.5 | 77.2 | |
| LlipPre-training Data=2.5B, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 89 | 95.7 | 81.4 | 70.9 | 88.2 | 41.5 | 63.7 | 93.5 | 94.7 | 74.9 | 75.3 | 79 | |
| MetaCLIPPre-training Data=2.5B, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 88.3 | 95.7 | 79 | 68.5 | 82.9 | 30.3 | 62.1 | 91.7 | 93.3 | 73.9 | 72.1 | 76.2 | |
| OpenCLIPPre-training Data=LAION-2B, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 86.6 | 94.9 | 76.9 | 70.5 | 88.5 | 26.8 | 56.5 | 90.2 | 90.9 | 71.4 | 70.2 | 74.9 | |
| OpenCLIPPre-training Data=LAION-400M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 86.1 | 91.7 | 71.4 | 69 | 83.7 | 17.9 | 50.7 | 89.3 | 90.1 | 69.3 | 67.1 | 71.5 | |
| DreamLIPPre-training Data=Merged-30M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 76.4 | 92.7 | 71.7 | 64.7 | 23.4 | 8.3 | 33.9 | 66.3 | 87.6 | 59.1 | 58.4 | 58.4 | |
| COSMOSPre-training Data=Merged-30M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 73.9 | 93.4 | 72.3 | 67.8 | 31.2 | 5.9 | 37.3 | 62.9 | 88 | 54.6 | 57.6 | 58.6 | |
| SF-CLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 65 | 85 | 53.6 | — | 8.2 | 11 | 35.2 | 38.1 | 72.2 | 59.5 | 45.2 | — | |
| MaskCLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 64.9 | 80.2 | 57.5 | — | 6.7 | 6.1 | 27.9 | 34.3 | 72 | 57 | 44.5 | — | |
| SigLIPPre-training Data=Merged-30M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 64.2 | 91 | 67.6 | 64 | 22 | 5.7 | 33.5 | 53.3 | 84.3 | 43.6 | 51 | 52.7 | |
| COSMOSPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 63.1 | 92.8 | 70.8 | 64.2 | 30.4 | 3.4 | 36.5 | 56.9 | 85.5 | 36.3 | 51.4 | 53.8 | |
| LaSLIPPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 62.9 | 82 | 50.2 | 59.6 | 32.2 | 4.4 | 30.1 | 70.6 | 82.4 | 37.4 | 49.7 | 51 | |
| CLIPPre-training Data=Merged-30M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 61.3 | 92.2 | 66.9 | 62.2 | 19.3 | 5.7 | 30.9 | 49.3 | 83.7 | 43.4 | 50 | 51.4 | |
| MLLM-APre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 60.9 | 83 | 55.4 | 59.4 | 24.1 | 3.2 | 30.7 | 64.8 | 79.3 | 36 | 47.5 | 49.5 | |
| LaCLIPPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 60.7 | 75.1 | 43.9 | 57 | 36.3 | 5.6 | 31 | 72.4 | 83.3 | 39.9 | 48.4 | 50.3 | |
| SLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 59.9 | 82.6 | 48.6 | — | 5.4 | 5.6 | 26.6 | 31.4 | 70.9 | 56.3 | 42.8 | — | |
| DreamLIPPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 59.2 | 87.7 | 63.8 | 54.5 | 29.9 | 5.1 | 29.5 | 60.8 | 82.5 | 32.5 | 50.2 | 50.5 | |
| COSMOSPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 55.8 | 90.4 | 65.6 | 59.7 | 16.8 | 5.8 | 33.7 | 39.3 | 83.1 | 46.9 | 52.4 | 50 | |
| SigLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 45.7 | 87.8 | 58.7 | 58.6 | 9.9 | 4.7 | 28.1 | 32.3 | 79.1 | 36.4 | 44.4 | 44.2 | |
| DreamLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 45 | 88.9 | 62.8 | 57.5 | 9.4 | 5.8 | 32 | 33.6 | 81.5 | 44.4 | 48.2 | 46.3 | |
| CLIPPre-training Data=YFCC15M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 43.5 | 88 | 56.8 | 55.8 | 9.3 | 4.7 | 26 | 27.9 | 76.7 | 37 | 41.7 | 42.5 | |
| SigLIPPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 42.3 | 88.9 | 59.6 | 57.6 | 11.2 | 2.3 | 24.1 | 38.4 | 80 | 13.5 | 38.9 | 41.5 | |
| CLIPPre-training Data=CC12M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 38.4 | 90.4 | 58.6 | 57.7 | 12.2 | 2 | 24.6 | 40.9 | 80.8 | 12.1 | 39.1 | 41.5 | |
| COSMOSPre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 31.2 | 79.1 | 51.6 | 54.2 | 5.2 | 2.3 | 27.4 | 37.5 | 78.1 | 25.6 | 37.1 | 39 | |
| DreamLIPPre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 23.6 | 75.7 | 44.2 | 46.1 | 3.6 | 1.6 | 18.7 | 27.4 | 73.6 | 18.5 | 31.6 | 33.1 | |
| MLLM-APre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 18.7 | 58.4 | 32.4 | 43.8 | 3.9 | 1.5 | 20.2 | 32.1 | 63.5 | 17.5 | 25 | 28.8 | |
| SigLIPPre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 18.4 | 76.4 | 41.9 | 46.9 | 3 | 1.4 | 17.6 | 20.6 | 70.4 | 10.8 | 25.9 | 30.3 | |
| CLIPPre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 17.9 | 75 | 40.8 | 43.1 | 2.6 | 1 | 15.3 | 22.1 | 68.9 | 12.6 | 23.9 | 29.4 | |
| LaCLIPPre-training Data=CC3M, Vision Encoder=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 14.2 | 57.1 | 27.5 | 35.1 | 1.6 | 1.6 | 16.6 | 15.6 | 52.7 | 14.7 | 21.5 | 23.5 |