Dimensional changes after CNN's convolution operation:
- Input dimensions:, respectively represent the length, width and height of the input sample
- Hyperparameters of convolution operation
- Number of convolution kernels:
- Convolution kernel size:
- Stride:
- Padding:
- The output dimension is , where
- Due to the parameter sharing mechanism of CNN, the number of parameters of each convolution kernel is , there are a total of weights and biases
- If you want to be consistent after convolution with a convolution matrix length and width before, when the time
- When the convolution kernel is 3, padding select 1
- When the convolution kernel is 5, padding select 2
- When the convolution kernel is 7, padding chooses 3
Dimension changes after CNN's pooling operation:
- Input dimensions:, respectively represent the length, width and height of the input sample
- Hyperparameters for pooling operations
- Pooling layer size:
- Stride:
- Padding:
- The output dimension is , where
Reference materials: https://blog.csdn.net/qq_41670192/article/details/79231732