Image Classification on ImageNet Variants and ObjectNet (test)
90.2IN-A AccuracyCoCa
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CoCazero-shot=true, Pre-training data=private JFT-3B dataset2023.12 | 90.2 | 86.3 | 96.5 | 80.7 | 77.6 | 82.7 | 0.6 | 85.7 | |
| LiT-22Bzero-shot=true, Pre-training data=private JFT-3B dataset2023.12 | 90.1 | 85.9 | 96 | 80.9 | — | 87.6 | — | — | |
| EVA-CLIP-18Bzero-shot=true2024.02 | 87.3 | 83.8 | 95.7 | 77.9 | 74.7 | 82.2 | 0.2 | 83.6 | |
| BASIC-Lzero-shot=true, reported=true2024.02 | 85.6 | 85.7 | 95.7 | 80.6 | 76.1 | 82.3 | 1.4 | 84.3 | |
| EVA-CLIP-8B (ours, distill)#Params=4.59B, FLOPs=2.35T, Throughout=30.1 imgs/s, Status=Distilled2026.03 | 85.6 | 83.7 | 95.3 | 77.7 | 74 | 81.3 | — | 82.9 | |
| EVA-CLIP-8Bzero-shot=true2024.02 | 85.2 | 83.5 | 95.3 | 77.7 | 74.3 | 81.2 | 0.6 | 82.9 | |
| EVA-CLIP-8B#Params=7.53B, FLOPs=3.86T, Throughout=20.8 imgs/s, Status=Original2026.03 | 85.2 | 83.5 | 95.3 | 77.7 | 74.3 | 81.2 | — | 82.9 | |
| InternVL-Czero-shot=true2023.12 | 83.8 | 83.2 | 95.5 | 77.3 | 73.9 | 80.6 | 0.8 | 82.4 | |
| InternVL-Czero-shot=true2024.02 | 83.8 | 83.2 | 95.7 | 77.3 | 74.3 | 80.6 | 0.7 | 82.5 | |
| SigLIP-SO+zero-shot=true, reported=true2024.02 | 82.5 | 83 | 95.8 | 77.2 | 74.5 | 77 | 1.3 | 81.7 | |
| EVA-CLIP-E (ours, distill)#Params=2.50B, FLOPs=1.28T, Throughout=69.5 imgs/s, Status=Distilled2026.03 | 82.3 | 82 | 94.6 | 75.1 | 71.7 | 79.6 | — | 81 | |
| EVA-02-CLIP-E+zero-shot=true2023.12 | 82.1 | 82 | 94.5 | 75.7 | 71.6 | 79.6 | 1.1 | 80.9 | |
| EVA-02-CLIP-E/14+zero-shot=true2024.02 | 82.1 | 82.1 | 94.7 | 75.7 | 72.2 | 79.6 | 1 | 81.1 | |
| EVA-CLIP-E#Params=4.35B, FLOPs=2.23T, Throughout=43.1 imgs/s, Status=Original2026.03 | 82.1 | 82 | 94.5 | 75.7 | 71.6 | 79.6 | — | 80.9 | |
| DFN5B-CLIP-H/14+zero-shot=true2024.02 | 79.6 | 84.3 | 93.6 | 78.3 | 73.3 | 79.6 | 2.8 | 81.5 | |
| OpenAI CLIP-L+zero-shot=true2023.12 | 77.5 | 76.6 | 89 | 70.9 | 61 | 72 | 2.1 | 74.5 | |
| EVA-01-CLIP-g+zero-shot=true2023.12 | 74.1 | 79.3 | 92.5 | 72.1 | 68.1 | 75.3 | 2.4 | 76.9 | |
| EVA-01-CLIP-g/14+zero-shot=true2024.02 | 74.1 | 79.3 | 92.7 | 72.5 | 68.4 | 75.3 | 2.2 | 77.1 | |
| EVA-01-CLIP-gzero-shot=true2023.12 | 73.6 | 78.5 | 92.5 | 71.5 | 67.3 | 72.3 | 2.5 | 76 | |
| EVA-01-CLIP-g/14zero-shot=true2024.02 | 73.6 | 78.5 | 92.5 | 71.5 | 67.6 | 72.3 | 2.5 | 76 | |
| SigLIP-SOzero-shot=true, reported=true2024.02 | 71.9 | 82 | 95.1 | 76.1 | 74 | 70.6 | 3.7 | 78.3 | |
| DFN5B-CLIP-H/14zero-shot=true2024.02 | 71.7 | 83.5 | 92.9 | 77.4 | 72.8 | 76.7 | 4.4 | 79.2 | |
| OpenCLIP-Gzero-shot=true2023.12 | 69.3 | 80.1 | 92.1 | 73.6 | 68.9 | 73 | 3.9 | 76.2 | |
| OpenCLIP-G/14zero-shot=true2024.02 | 69.3 | 80.4 | 92.8 | 73.6 | 69.9 | 73 | 3.9 | 76.5 | |
| OpenCLIP-G#Params=1.84B, FLOPs=0.95T, Throughout=90.6 imgs/s, Status=Original2026.03 | 69.3 | 80.1 | 92.1 | 73.6 | 68.9 | 73 | — | 76.2 | |
| OpenCLIP-G (ours, distill)#Params=1.17B, FLOPs=0.60T, Throughout=135.8 imgs/s, Status=Distilled2026.03 | 68.4 | 80.1 | 91.9 | 73.4 | 68.9 | 72.8 | — | 75.9 | |
| OpenCLIP-gzero-shot=true2023.12 | 60.8 | 78.5 | 90.2 | 71.7 | 67.5 | 69.2 | 5.5 | 73 | |
| OpenCLIP-g#Params=1.01B, FLOPs=0.52T, Throughout=150.8 imgs/s, Status=Original2026.03 | 60.8 | 78.5 | 90.2 | 71.7 | 67.5 | 69.2 | — | 73 | |
| OpenCLIP-g (ours, distill)#Params=0.62B, FLOPs=0.32T, Throughout=222.3 imgs/s, Status=Distilled2026.03 | 60.8 | 78.4 | 90.5 | 71.5 | 67.9 | 69.4 | — | 73.1 | |
| OpenCLIP-Hzero-shot=true2023.12 | 59.3 | 78 | 89.3 | 70.9 | 66.6 | 69.7 | 5.7 | 72.3 | |
| EVA-CLIP-8B (ours, prune)#Params=4.59B, FLOPs=2.35T, Throughout=30.1 imgs/s, Status=Pruned2026.03 | 49.9 | 61.7 | 66.5 | 53.1 | 34.9 | 47.9 | — | 52.3 | |
| CLIP-ELearning Protocol=Transductive Learning2024.11 | 45.42 | — | 71.1 | — | 77.08 | — | — | 64.53 | |
| UMFC + CLIP-ELearning Protocol=Transductive Learning2024.11 | 44.77 | — | 72.19 | — | 78.62 | — | — | 65.19 | |
| UMFCLearning Protocol=Transductive Learning2024.11 | 43.42 | — | 68.86 | — | 77.24 | — | — | 63.17 | |
| CLIPLearning Protocol=Transductive Learning2024.11 | 42.13 | — | 66.95 | — | 74.58 | — | — | 61.22 | |
| OpenCLIP-G (ours, prune)#Params=1.17B, FLOPs=0.60T, Throughout=135.8 imgs/s, Status=Pruned2026.03 | 37.1 | 70.2 | 78.3 | 62.7 | 50.9 | 51.1 | — | 58.4 | |
| OpenCLIP-g (ours, prune)#Params=0.62B, FLOPs=0.32T, Throughout=222.3 imgs/s, Status=Pruned2026.03 | 34 | 65 | 74.4 | 57.7 | 43.9 | 47.9 | — | 53.8 | |
| EVA-CLIP-E (ours, prune)#Params=2.50B, FLOPs=1.28T, Throughout=69.5 imgs/s, Status=Pruned2026.03 | 18.6 | 59.2 | 59.8 | 51.3 | 34.9 | 40.6 | — | 44.1 | |
| MAWS-ViT-2Bzero-shot=true2023.12 | — | 81.9 | — | — | — | — | — | — |