Fine-grained classification on ImageNet (novel classes)
1.49ECEPenalty
Evaluation results
| Method | Links | ||
|---|---|---|---|
| PenaltyPrompt-tuning strategy=MaPLe, Calibration strategy=Penalty2026.02 | 1.49 | 70.66 | |
| TCPTPrompt-tuning strategy=MaPLe, Calibration strategy=TCPT2026.02 | 1.74 | 70.28 | |
| ZS-NormPrompt-tuning strategy=KGCoOp, Calibration strategy=ZS-Norm2026.02 | 1.8 | 69.68 | |
| PenaltyPrompt-tuning strategy=KGCoOp, Calibration strategy=Penalty2026.02 | 1.82 | 69.58 | |
| KGCoOpPrompt-tuning strategy=KGCoOp, Calibration strategy=None2026.02 | 1.84 | 69.7 | |
| TCPTPrompt-tuning strategy=KGCoOp, Calibration strategy=TCPT2026.02 | 1.84 | 69.5 | |
| MaPLePrompt-tuning strategy=MaPLe, Calibration strategy=None2026.02 | 1.93 | 70.5 | |
| Temp. ScalingPrompt-tuning strategy=MaPLe, Calibration strategy=Temp. Scaling2026.02 | 1.95 | 70.46 | |
| TCPTPrompt-tuning strategy=CoOp, Calibration strategy=TCPT2026.02 | 2.02 | 67.03 | |
| Zero ShotPrompt-tuning strategy=None, Calibration strategy=None2026.02 | 2.09 | 72.4 | |
| DACPrompt-tuning strategy=MaPLe, Calibration strategy=DAC2026.02 | 2.11 | — | |
| PenaltyPrompt-tuning strategy=CoOp, Calibration strategy=Penalty2026.02 | 2.36 | 66.71 | |
| ZS-NormPrompt-tuning strategy=CoOp, Calibration strategy=ZS-Norm2026.02 | 2.46 | 66.26 | |
| ZS-NormPrompt-tuning strategy=MaPLe, Calibration strategy=ZS-Norm2026.02 | 3.67 | 70.63 | |
| MBLSPrompt-tuning strategy=CoOp, Calibration strategy=MBLS2026.02 | 4.09 | 59.11 | |
| DACPrompt-tuning strategy=KGCoOp, Calibration strategy=DAC2026.02 | 4.32 | — | |
| MBLSPrompt-tuning strategy=KGCoOp, Calibration strategy=MBLS2026.02 | 4.6 | 69.14 | |
| DACPrompt-tuning strategy=CoOp, Calibration strategy=DAC2026.02 | 5.67 | — | |
| Temp. ScalingPrompt-tuning strategy=KGCoOp, Calibration strategy=Temp. Scaling2026.02 | 5.81 | 69.79 | |
| Temp. ScalingPrompt-tuning strategy=CoOp, Calibration strategy=Temp. Scaling2026.02 | 7.33 | 59.07 | |
| CoOpPrompt-tuning strategy=CoOp, Calibration strategy=None2026.02 | 10.69 | 59.07 | |
| MBLSPrompt-tuning strategy=MaPLe, Calibration strategy=MBLS2026.02 | 22.82 | 68.47 |