Low-shot Image Classification on iNaturalist-18 (val)
73.7Top-1 Accuracy (5-shot)MAWS
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MAWSPre-training Dataset=IG-3B, Architecture=ViT-6.5B2023.03 | 73.7 | 36.4 | 80.9 | |
| MAWSPre-training Dataset=IG-3B, Architecture=ViT-2B2023.03 | 72.8 | 35.5 | 80.3 | |
| MAWSPre-training Dataset=IG-3B, Architecture=ViT-H2023.03 | 67.8 | 31.7 | 76.1 | |
| SWAGPre-training Dataset=IG-3.6B, Architecture=ViT-H2023.03 | 62.8 | 30.1 | 72.3 | |
| OpenCLIPPre-training Dataset=LAION-2B, Architecture=ViT-G2023.03 | 55.7 | 26.3 | 65.1 | |
| OpenCLIPPre-training Dataset=LAION-2B, Architecture=ViT-H2023.03 | 54.6 | 26 | 63.7 | |
| CLIPPre-training Dataset=WIT-400M, Architecture=ViT-L2023.03 | 49 | 21.9 | 58.5 | |
| MAEPre-training Dataset=IN1k, Architecture=ViT-H2023.03 | 48.5 | — | 68.2 | |
| DINOPre-training Dataset=IN1k, Architecture=ViT-B/82023.03 | 45.9 | 19.8 | 55.9 | |
| MSNPre-training Dataset=IN1k, Architecture=ViT-L/72023.03 | 38 | 17 | 48.1 |