Base-to-New Classification on 11 downstream datasets Balanced, τ=1
73.92IN AccuracyCPT
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| CPTBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 73.92 | 96.53 | 96.37 | 77.21 | 84.69 | 91.17 | 40.13 | 80.5 | 71.2 | 87.54 | 83.83 | 80.28 | |
| DPCBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 73.85 | 96.5 | 96.71 | 78.46 | 83.59 | 91.49 | 40.24 | 80.76 | 66.84 | 84.13 | 83.21 | 79.62 | |
| NPTBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 73.82 | 96.23 | 96.24 | 72.49 | 82.37 | 91.67 | 36.46 | 79.76 | 68.88 | 82.12 | 81.42 | 78.31 | |
| MaPLeBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 73.47 | 96.02 | 96.28 | 73.47 | 82.56 | 91.38 | 36.5 | 79.75 | 68.16 | 82.35 | 80.77 | 78.25 | |
| DeKgBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 73.07 | 96.89 | 96.09 | 77.83 | 84.51 | 91.14 | 39.51 | 80.35 | 69.7 | 87.42 | 84.8 | 80.12 | |
| CoPromptBackbone=ViT-B/16, Training protocol=trained from base classes, Max samples per class (max{nk})=16, Evaluation Metric=Harmonic Mean (H)2026.05 | 72.86 | 95.94 | 96.33 | 75.47 | 84.51 | 90.95 | 37.82 | 79.29 | 68.93 | 83.07 | 80.78 | 78.72 |