Page 82 - Read Online

Bah et al. Intell Robot 2022;2(1):72­88 I http://dx.doi.org/10.20517/ir.2021.16 Page 76

Table 1. Architecture detail
Layer (type) Output shape Param #
input (InputLayer) [(None, 48, 48, 1)] 0
conv2d_1 (Conv2D) (None, 48, 48, 16) 416
batch_normalization_1 (BatchNormalization) (None, 48, 48, 16) 64
conv2d_2 (Conv2D) (None, 48, 48, 16) 6416
batch_normalization_2 (BatchNormalization) (None, 48, 48, 16) 64
dropout_1 (Dropout) (None, 48, 48, 16) 0
conv2d_3 (Conv2D) (None, 48, 48, 32) 4640
batch_normalization_3 (BatchNormalization) (None, 48, 48, 32) 128
conv2d_4 (Conv2D) (None, 48, 48, 32) 9248
batch_normalization_4 (BatchNormalization) (None, 48, 48, 32) 128
max_pooling2d_1 (MaxPooling2D) (None, 24, 24, 32) 0
dropout_2 (Dropout) (None, 24, 24, 32) 0
conv2d_5 (Conv2D) (None, 24, 24, 32) 9248
batch_normalization_5 (BatchNormalization) (None, 24, 24, 32) 128
activation_1 (Activation) (None, 24, 24, 32) 0
conv2d_6 (Conv2D) (None, 24, 24, 32) 9248
batch_normalization_6 (BatchNormalization) (None, 24, 24, 32) 128
add_1 (Add) (None, 24, 24, 32) 0
activation_2 (Activation) (None, 24, 24, 32) 0
conv2d_7 (Conv2D) (None, 24, 24, 64) 18496
batch_normalization_7 (BatchNormalization) (None, 24, 24, 64) 256
conv2d_8 (Conv2D) (None, 24, 24, 64) 36928
batch_normalization_8 (BatchNormalization) (None, 24, 24, 64) 256
max_pooling2d_2 (MaxPooling2D) (None, 12, 12, 64) 0
dropout_3 (Dropout) (None, 12, 12, 64) 0
conv2d_9 (Conv2D) (None, 12, 12, 128) 73856
batch_normalization_9 (BatchNormalization) (None, 12, 12, 128) 512
conv2d_10 (Conv2D) (None, 12, 12, 128) 147584
batch_normalization_10 (BatchNormalization) (None, 12, 12, 128) 512
max_pooling2d_3 (MaxPooling2D) (None, 6, 6, 128) 0
dropout_4 (Dropout) (None, 6, 6, 128) 0
conv2d_11 (Conv2D) (None, 6, 6, 128) 147584
batch_normalization_11 (BatchNormalization) (None, 6, 6, 128) 512
activation_3 (Activation) (None, 6, 6, 128) 0
conv2d_12 (Conv2D) (None, 6, 6, 128) 147584
batch_normalization_12 (BatchNormalization) (None, 6, 6, 128) 512
add_2 (Add) (None, 6, 6, 128) 0
activation_4 (Activation) (None, 6, 6, 128) 0
conv2d_13 (Conv2D) (None, 6, 6, 256) 295168
batch_normalization_13 (BatchNormalization) (None, 6, 6, 256) 1024
conv2d_14 (Conv2D) (None, 6, 6, 256) 590080
batch_normalization_14 (BatchNormalization) (None, 6, 6, 256) 1024
max_pooling2d_4 (MaxPooling2D) (None, 3, 3, 256) 0
dropout_5 (Dropout) (None, 3, 3, 256) 0
conv2d_15 (Conv2D) (None, 3, 3, 512) 1180160
batch_normalization_15 (BatchNormalization) (None, 3, 3, 512) 2048
conv2d_16 (Conv2D) (None, 3, 3, 512) 2359808
batch_normalization_16 (BatchNomalization) (None, 3, 3, 512) 2048
max_pooling2d_5 (MaxPooling2D) (None, 1, 1, 512) 0
dropout_6 (Dropout) (None, 1, 1, 512) 0
conv2d_17 (Conv2D) (None, 1, 1, 512) 2359808
batch_normalization_17 (BatchNormalization) (None, 1, 1, 512) 2048
activation_5 (Activation) (None, 1, 1, 512) 0
conv2d_18 (Conv2D) (None, 1, 1, 512) 2359808
batch_normalization_18 (BatchNormalization) (None, 1, 1, 512) 2048
add_3 (Add) (None, 1, 1, 512) 0
activation_6 (Activation) (None, 1, 1, 512) 0
global_average_pooling2d_1 (GlobalAveragePooling2D) (None, 512) 0
dense_1 (Dense) (None, 7) 3591
activation_7 (Activation) (None, 7) 0
Total params: 9,773,111
Trainable params: 9,766,391
Non-trainable params: 6,720

formula:
− + 2
= + 1 (1)