Image Classification on CLIP Zero-shot Evaluation Suite (10 datasets)
89.6Cars AccuracyCLIP-L
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CLIP-LTraining Data=LAION400M, Model Architecture=ViT-L/14, Evaluation Protocol=Zero-shot2024.12 | 89.6 | 90.1 | 94.6 | 77.4 | 72.6 | 25 | 60.4 | 91.7 | 82.1 | 75.5 | 75.9 | |
| CLIP-BTraining Data=LAION400M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 83.5 | 85.5 | 93 | 71.7 | 66.8 | 16.7 | 52.8 | 90.1 | 91.2 | 63.9 | 65.5 | |
| SAIL-L-NV2Training Data=23M Merged, Model Architecture=ViT-L/14, Evaluation Protocol=Zero-shot2024.12 | 44.6 | 86.1 | 96.7 | 86.7 | 69.8 | 28.6 | 63.5 | 82.3 | 85.4 | 77.2 | 72.1 | |
| SAIL-L-NV2Training Data=CC12M, Model Architecture=ViT-L/14, Evaluation Protocol=Zero-shot2024.12 | 42.9 | 81.9 | 96.1 | 85.2 | 68.3 | 16.3 | 60.4 | 84.7 | 82.4 | 67.5 | 68.6 | |
| SAIL-B-NV2Training Data=CC12M, Model Architecture=ViT-B/16, Patch Size=14, Evaluation Protocol=Zero-shot2024.12 | 35.8 | 77.7 | 93.8 | 79.9 | 66.2 | 13.4 | 61.5 | 81.7 | 82.1 | 61.5 | 65.4 | |
| SAIL-L-GTETraining Data=CC12M, Model Architecture=ViT-L/14, Evaluation Protocol=Zero-shot2024.12 | 33 | 71.2 | 96.3 | 83.8 | 67.2 | 8 | 53 | 66.5 | 82.6 | 57.7 | 61.9 | |
| SAIL-L-GTETraining Data=23M Merged, Model Architecture=ViT-L/14, Evaluation Protocol=Zero-shot2024.12 | 32 | 76.1 | 97.3 | 84.6 | 68.6 | 16 | 52.5 | 56.9 | 83 | 68.3 | 63.5 | |
| DreamLIPTraining Data=CC12M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 29.7 | 58.3 | 87.3 | 62.6 | 54.3 | 4.9 | 29.2 | 60.3 | 83.1 | 28.9 | 49.9 | |
| SAIL-B-GTETraining Data=CC12M, Model Architecture=ViT-B/16, Patch Size=14, Evaluation Protocol=Zero-shot2024.12 | 28.1 | 63.1 | 94.1 | 78.2 | 64.2 | 6.6 | 52 | 60.1 | 81.5 | 49.4 | 57.7 | |
| CLIP-BTraining Data=CC12M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 14.7 | 25.3 | 66.5 | 32.1 | 39.9 | 1.9 | 13.5 | 45 | 59.8 | 15 | 31.4 | |
| LiTTraining Data=CC12M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 13.2 | — | — | — | — | 5 | 74.4 | — | — | 48.2 | — | |
| ShareLock+Training Data=CC12M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 11.5 | — | — | — | — | 8.3 | 66.6 | — | — | 48.8 | — | |
| SAIL-B-GTETraining Data=CC3M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 9.2 | 47.1 | 94.1 | 74.6 | 63.9 | 4.2 | 49.7 | 39.5 | 77.9 | 31.8 | 49.2 | |
| LiTTraining Data=CC3M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 3 | — | — | — | — | 2.1 | 28.5 | — | — | 35.9 | — | |
| DreamLIPTraining Data=CC3M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 2.8 | 19.4 | 74.3 | 44.2 | 45.9 | 1 | 17 | 27.1 | 63.1 | 14.7 | 31 | |
| CLIP-BTraining Data=CC3M, Model Architecture=ViT-B/16, Evaluation Protocol=Zero-shot2024.12 | 1.2 | 10.6 | 53.9 | 20.4 | 31.2 | 1.1 | 10.4 | 11.7 | 43.2 | 12.9 | 19.7 |