Few-shot Classification on ImageNet-1K 16-shot
70.0216-shot AccuracyAMU-Tuning
Evaluation results
| Method | Links | |
|---|---|---|
| AMU-TuningBias=W f^Aux, Feature=Aux, Predictor=MTFi LP, Fusion=Uncertainty-based2024.04 | 70.02 | |
| CaFoBias=alpha phi(F_CLIP^c f_V^T)V + (1-alpha)phi(F_DINO^c f_V^T)V, Feature=CLIP+DINO, Predictor=Cache, Fusion=Similarity-based2024.04 | 68.79 | |
| Tip-Adapter-FBias=phi(F^c f_V^T)V, Feature=CLIP, Predictor=Cache, Fusion=Manual Tuning2024.04 | 65.51 | |
| CLIP-AdapterBias=f_T^ada(W_0)f_V^T + W_0 f_V^ada(f_V^T) + f_T^ada(W_0)f_V^ada(f_V^T), Feature=CLIP, Predictor=MLP, Fusion=Manual Tuning2024.04 | 63.59 | |
| CoOpBias=f_T(T_bias)f_V^T, Feature=T_bias, Predictor=f_T, Fusion=-2024.04 | 62.95 | |
| Zero-shot CLIPBias=-, Feature=-, Predictor=-, Fusion=-2024.04 | 60.33 |