Image Classification on ImageNet-1K (val) (kNN Accuracy)
86.1kNN AccuracyEVA-CLIP-8B
Evaluation Results
| Method | Links | |
|---|---|---|
| EVA-CLIP-8BMode=distill, #Params=4.59B, FLOPs=2.35T2026.03 | 86.1 | |
| EVA-CLIP-8BSource=Sun et al., 2024, #Params=7.53B, FLOPs=3.86T2026.03 | 86 | |
| EVA-CLIP-EMode=distill, #Params=2.50B, FLOPs=1.28T2026.03 | 85.9 | |
| EVA-CLIP-ESource=Sun et al., 2023, #Params=4.35B, FLOPs=2.23T2026.03 | 85.8 | |
| DINOv2-gSource=Oquab et al., 2023, #Params=1.14B, FLOPs=0.30T2026.03 | 83.5 | |
| DINOv2-gMode=distill, #Params=0.62B, FLOPs=0.16T2026.03 | 83.5 | |
| OpenCLIP-GSource=Wortsman, 2023, #Params=1.84B, FLOPs=0.95T2026.03 | 82.9 | |
| OpenCLIP-GMode=distill, #Params=1.17B, FLOPs=0.60T2026.03 | 82.8 | |
| OpenCLIP-gSource=Cherti et al., 2023, #Params=1.01B, FLOPs=0.52T2026.03 | 81.7 | |
| OpenCLIP-gMode=distill, #Params=0.62B, FLOPs=0.32T2026.03 | 81.7 | |
| DINOv2-gMode=prune, #Params=0.62B, FLOPs=0.16T2026.03 | 76.4 | |
| OpenCLIP-GMode=prune, #Params=1.17B, FLOPs=0.60T2026.03 | 74.7 | |
| EVA-CLIP-8BMode=prune, #Params=4.59B, FLOPs=2.35T2026.03 | 74.4 | |
| OpenCLIP-gMode=prune, #Params=0.62B, FLOPs=0.32T2026.03 | 72.8 | |
| EVA-CLIP-EMode=prune, #Params=2.50B, FLOPs=1.28T2026.03 | 72.2 |