Table 2 50-way classification accuracy on CUB
From: Multimodal few-shot classification without attribute embedding
Method | Metric | 1-shot | 2-shot | 5-shot | 10-shot | 20-shot |
|---|---|---|---|---|---|---|
With attribute embedding | ||||||
Pahde et al. [26] | Top-1 | 24.90 | 25.17 | 34.66 | 44.00 | 53.70 |
Top-3 | 37.59 | 39.75 | 49.86 | 59.62 | 67.99 | |
Top-5 | 57.67 | 59.83 | 73.01 | 78.10 | 84.24 | |
Multimodal prototypical | Top-1 | 34.16 | 41.43 | 48.84 | 53.01 | 55.58 |
Network [8] | Top-3 | 58.56 | 67.44 | 74.65 | 77.60 | 79.30 |
Top-5 | 70.39 | 78.62 | 84.32 | 86.23 | 87.47 | |
Without attribute embedding | ||||||
Proposed method (ResNet-18) | Top-1 | 29.00 | 33.83 | 49.10 | 57.60 | 64.57 |
Top-3 | 49.74 | 53.89 | 70.96 | 80.83 | 84.70 | |
Top-5 | 59.89 | 64.74 | 80.79 | 87.92 | 90.59 | |
Proposed method (ResNet-101) | Top-1 | 42.61 | 52.95 | 63.67 | 71.79 | 74.73 |
Top-3 | 66.30 | 75.76 | 85.57 | 88.16 | 90.70 | |
Top-5 | 73.57 | 84.90 | 90.98 | 94.00 | 94.36 | |