Image Classification on ImageNet H (harmonic mean)
74.53AccuracyFARL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FARLProtocol=Base-to-novel generalization2025.12 | 74.53 | — | |
| MMRLProtocol=Base-to-novel generalization2025.12 | 74.37 | — | |
| MetaPromptProtocol=Base-to-novel generalization2025.12 | 74.02 | — | |
| MMAProtocol=Base-to-novel generalization2025.12 | 74.02 | — | |
| MaPLeProtocol=Base-to-novel generalization2025.12 | 73.47 | — | |
| PLOTProtocol=Base-to-novel generalization2025.12 | 73.4 | — | |
| TCPProtocol=Base-to-novel generalization2025.12 | 73.38 | — | |
| CoCoOpProtocol=Base-to-novel generalization2025.12 | 73.1 | — | |
| KgCoOpProtocol=Base-to-novel generalization2025.12 | 72.78 | — | |
| ProVPProtocol=Base-to-novel generalization2025.12 | 72.36 | — | |
| CoOpProtocol=Base-to-novel generalization2025.12 | 71.02 | — | |
| CLIPProtocol=Base-to-novel generalization2025.12 | 70.22 | — | |
| CoCoOpBackbone=ViT-B/16, Shots=162023.11 | 0.731 | — | |
| BadCLIPBackbone=ViT-B/16, Shots=16, Context length=4, Maximum noise strength (epsilon)=42023.11 | 0.729 | 99.65 | |
| CoOpBackbone=ViT-B/16, Shots=162023.11 | 0.7192 | — | |
| CLIPBackbone=ViT-B/16, mode=zero-shot2023.11 | 0.7022 | — |