Human-Computer Interaction with Hand Gesture Recognition Using ResNet and MobileNet

Table 2

Summary of the modified ResNet50 model.

Transfer learning	Pretrained on Google ImageNet dataset
Number of layers	50 layer of ResNet50 architecture +3 sequential layers (2 fully connected layers + 1 softmax layer). Therefore, the total number of layers is 53 layers
Number of neurons in fully connected layers	1,024 neurons in the 1 fully connected layer, 512 neurons in the 2nd fully connected layer, and 32 neurons in the last layer (softmax layer)
Activation function	ReLu activation function
Optimizer	Adam optimizer with 0.0001 learning rate
Dropout	Dropout value used for the fully connected layers is 0.5
Batch size	32
Loss function	Categorical loss entropy
Number of epochs	10 epoch
Total no. of parameters	Total params: 32,495,648- trainable params: 32,450,208- nontrainable params: 45,440