Few-shot Classification on DTD 16-shot (test)
63.73Accuracy (Sym 0.125)NA-MVP
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| NA-MVPImage Encoder=ResNet-50, Text Encoder=63M-parameter Transformer, Training Shots=16, Training Epochs=502026.03 | 63.73 | 63.13 | 61.63 | 58.5 | 52.93 | 48.63 | 62.33 | 52.1 | |
| NLPromptImage Encoder=ResNet-50, Text Encoder=63M-parameter Transformer, Training Shots=16, Training Epochs=502026.03 | 62.97 | 61.23 | 59.17 | 55.17 | 49.03 | 39.8 | 60.6 | 50.8 | |
| GCEImage Encoder=ResNet-50, Text Encoder=63M-parameter Transformer, Training Shots=16, Training Epochs=502026.03 | 61 | 59.83 | 56.8 | 50.73 | 43.6 | 33.67 | 57.57 | 43.97 | |
| CoOpImage Encoder=ResNet-50, Text Encoder=63M-parameter Transformer, Training Shots=16, Training Epochs=502026.03 | 56 | 49.57 | 43.3 | 34.37 | 27.83 | 17.27 | 47.75 | 29.63 | |
| JoAPRImage Encoder=ResNet-50, Text Encoder=63M-parameter Transformer, Training Shots=16, Training Epochs=502026.03 | 55.02 | 53.95 | 51.57 | 49.12 | 44.24 | 35.9 | 49.23 | 38.33 |