Table 2 The details of hyper-parameters applied for various CNN models
CNN model | Batch-size | Learning-rate |
|---|---|---|
AlexNet | 48,24,16 | 8e\(-\)7,1e\(-\)8,1e\(-\)7,6e\(-\)7,1e\(-\)5,1e\(-\)6,1e\(-\)10 |
SqueezeNet | 48,24,16 | 1e\(-\)8, 1e\(-\)7, 1e\(-\)6, 9e\(-\)5 |
GoogleNet | 48,24,16 | 1e\(-\)8, 6e\(-\)7, 1e\(-\)7, 9e\(-\)6, 1e\(-\)6, 1e\(-\)5 |
MobileNetV2 | 24,16,8 | 9e\(-\)6, 3e\(-\)6,5e\(-\)6,1e\(-\)7,8e\(-\)5,1e\(-\)6 |
InceptionV3 | 48,24,16 | 8e\(-\)5,1e\(-\)6,9e\(-\)6,1e\(-\)5,1e\(-\)7 |
Our ensemble model (MobileNetV2 +InceptionV3 ) | 24,16,8 | 4e-7, 1e-7, 1e-5, 1e-6 |