Table 3 5-way classification accuracy of Top-1 on CUB
From: Multimodal few-shot classification without attribute embedding
Method | 1-shot | 5-shot |
|---|---|---|
With attribute embedding | ||
AM3-TADAM [27] | 74.10 | 79.70 |
Multimodal prototypical network [8] | 75.01 | 85.30 |
Dual TriNet [28] | 69.61 | 84.10 |
f-VAEGAN-D2 [13] | 84.00 | 85.00 |
RS-FSL [38] | 65.66 | – |
L3 [29] | 53.96 | – |
LSL [30] | 61.24 | – |
Multiple-semantics [25] | 76.10 | 82.90 |
Without attribute embedding | ||
Proposed method (ResNet-18) | 64.68 | 85.36 |
Proposed method (ResNet-101) | 74.67 | 87.22 |