Image Classification on ImageNet and ObjectNet Robustness Suite
80.9Average AccuracyEVA-02-CLIP-E/14+
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| EVA-02-CLIP-E/14+zero-shot=true2023.03 | 80.9 | 1.1 | — | — | — | — | — | — | |
| EVA-02-CLIP-L/14+zero-shot=true2023.03 | 79.6 | 0.8 | — | — | — | — | — | — | |
| DynamiCSDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=2.56B@112 + 128M@224, GPU-hours=299, Zero-shot=true2026.04 | 59.1 | — | 37.2 | 57.6 | 79 | 57.4 | 64.1 | 59.3 | |
| DynamiCSDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=1.28B@112 + 128M@224, GPU-hours=163, Zero-shot=true2026.04 | 58.2 | — | 35.7 | 58.4 | 76.6 | 55.9 | 64 | 58.4 | |
| MetaCLIP-400MDataset (Data Size)=400M, Samples Seen@Resolution=12.8B@224, GPU-hours=≈ 10700, Zero-shot=true2026.04 | 57.5 | — | — | — | — | — | — | — | |
| HQ-CLIPDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=3.20B@224, GPU-hours=≈ 2675, Zero-shot=true2026.04 | 57.2 | — | 39.1 | 43 | 80.1 | 57.3 | 63.1 | 60.6 | |
| DFN*Dataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=1.28B@112 + 128M@224, GPU-hours=151, Zero-shot=true2026.04 | 56 | — | 31.4 | 59.7 | 74.5 | 53.9 | 60.9 | 55.5 | |
| OpenAI-WITDataset (Data Size)=400M, Samples Seen@Resolution=12.8B@224, GPU-hours=10700, Zero-shot=true2026.04 | 55.9 | — | 50 | 42.3 | 77.7 | 48.2 | 55.3 | 61.9 | |
| DynamiCSDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=0.64B@112 + 128M@224, GPU-hours=95, Zero-shot=true2026.04 | 55.7 | — | 30.4 | 62.3 | 73.2 | 53.2 | 60.7 | 54.6 | |
| OpenCLIPDataset (Data Size)=LAION-400M (400M), Samples Seen@Resolution=12.8B@224, GPU-hours=10736, Zero-shot=true2026.04 | 54.1 | — | 33.2 | 50.8 | 77.9 | 52.4 | 50.8 | 59.6 | |
| DFNDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=1.28B@224, GPU-hours=≈ 1070, Zero-shot=true2026.04 | 54 | — | — | — | — | — | — | — | |
| DynamiCSDataset (Data Size)=LAION-400M (298M), Samples Seen@Resolution=2.56B@112 + 128M@224, GPU-hours=299, Zero-shot=true2026.04 | 53.7 | — | 30.9 | 53.5 | 76.3 | 51.7 | 59.2 | 50.7 | |
| DynamiCSDataset (Data Size)=LAION-400M (298M), Samples Seen@Resolution=1.28B@112 + 128M@224, GPU-hours=163, Zero-shot=true2026.04 | 52 | — | 28.9 | 56.1 | 73.2 | 49.2 | 56.4 | 47.9 | |
| RandomDataset (Data Size)=DataComp-DFN (130M), Samples Seen@Resolution=0.64B@112 + 128M@224, GPU-hours=90, Zero-shot=true2026.04 | 51.7 | — | 23 | 59.5 | 70.3 | 49.3 | 56.4 | 51.4 | |
| CLIPA*Dataset (Data Size)=LAION-400M (298M), Samples Seen@Resolution=2.56B@112 + 128M@224, GPU-hours=269, Zero-shot=true2026.04 | 50.9 | — | 26.7 | 54.1 | 73.3 | 48.6 | 55.5 | 47.2 | |
| DataCompDataset (Data Size)=DataComp-Large (1.28B), Samples Seen@Resolution=1.28B@224, GPU-hours=≈ 1070, Zero-shot=true2026.04 | 50.8 | — | 25.5 | 49.6 | 71.8 | 49.8 | 55.1 | 53.1 | |
| RECLIP*Dataset (Data Size)=LAION-400M (298M), Samples Seen@Resolution=2.56B@112 + 128M@224, GPU-hours=280, Zero-shot=true2026.04 | 50.5 | — | 26.1 | 53.5 | 72.8 | 48.1 | 55.2 | 47.2 | |
| RandomDataset (Data Size)=LAION-400M (298M), Samples Seen@Resolution=1.28B@112 + 128M@224, GPU-hours=151, Zero-shot=true2026.04 | 47.4 | — | 21.6 | 53.7 | 69.3 | 44.9 | 52 | 43.1 |