Zero-shot Image Classification on ImageNet zero-shot (Loss and Uncertainty)
0.942Average LossPre-trained
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Pre-trainedModel=CLIP-VIT-G/14, # Params=1.01B, System=FP322024.11 | 0.942 | — | 0.942 | 0.942 | 20.6 | 9.5 | |
| Pre-trainedModel=CLIP-VIT-L/14, # Params=0.30B, System=FP322024.11 | 0.948 | — | 0.948 | 0.948 | 25.1 | 4.9 | |
| LPE-BSRModel=CLIP-VIT-G/14, # Params=1.01B, System=INT-5, Ensemble size=202024.11 | 0.955 | 0.013 | 0.941 | 0.927 | 20.6 | 8.9 | |
| LPE-BSRModel=CLIP-VIT-L/14, # Params=0.30B, System=INT-5, Ensemble size=202024.11 | 0.993 | 0.053 | 0.94 | 0.929 | 25 | 2.8 |