| parameter | |||||
|---|---|---|---|---|---|
| model_type | ConvNet4 | ConvNet6 | ResNet10 | ResNet18 | ResNet34 |
| dataset | CUB | MiniImageNet | |||
| if_augmentation | True | False | |||
| num_epochs | 600 | ||||
| batch_size | 1 | ||||
| sgd_lr | 0.1 | ||||
| num_way | 5 | ||||
| num_shot | 5 | 1 | |||
| num_query | 16 | ||||
| num_train | 100 | ||||
| num_val | 600 | ||||
| num_test | 600 | ||||
| proxy_type | Proxy | Mean | Sum | ||
| Classifier | 3DConv | FC | Euclidean |
Table:The performance of ProxyNet with data augmentation
| CUB | CUB | mini-ImageNet | mini-ImageNet | |
|---|---|---|---|---|
| Embedding Network | 1-shot | 5-shot | 1-shot | 5-shot |
| Conv-4 | 67.52±0.97 | 82.85±0.60 | 52.95±0.76 | 70.35±0.63 |
| Conv-6 | 68.16±0.93 | 83.57±0.58 | 52.18±0.82 | 69.91±0.62 |
| ResNet-10 | 76.79±0.84 | 88.02±0.52 | 58.16±0.87 | 75.27±0.65 |
| ResNet-18 | 76.72±0.90 | 88.63±0.49 | 57.88±0.87 | 75.23±0.66 |
| ResNet-34 | 77.70±0.86 | 87.05±0.52 | 58.54±0.85 | 75.90±0.61 |
Reference Code: FEAT https://github.com/Sha-Lab/FEAT