Table 5 Test accuracy results from using different backbone variations in EPNet on and testing on mini-ImageNet, BCCD, and HEp-2.
From: Automated human cell classification in sparse datasets using few-shot learning
Backbone (%) | Mini-ImageNet (%) | BCCD | HEp-2 (%) |
|---|---|---|---|
Backbone performance | |||
WideResNet28-10 (Original Backbone) | 88.7 | 47.4 | 55.1 |
EfficientNetV2 (Default Width) | 59.8 | 18.3 | 26.3 |
EfficientNetV2 (0.5 Width) | 67.3 | 25.7 | 33.3 |
EfficientNetV2 (0.75 Width) | 69.2 | 28.0 | 35.7 |
EfficientNetV2 (2.75 Width) | 70.8 | 29.5 | 37.1 |
ResNet-18 | 68.2 | 26.8 | 34.7 |
DenseNet | 78.8 | 37.4 | 45.0 |
