The calculation of stripe, kernel and padding in CNN

During the construction of convolution network , Encountered a confusion , That is how to calculate the output size of the network according to some super parameters of convolution , Convolution size change

In order to make the calculation process more intuitive , Define the following parameters

Definition

Definition  

Definition

Enter the dimension as

The output size is

Then the convolution size changes to  

for example , Input is 28  , Convolution kernels for 3*3  , step stride=1,padding=1, From the above formula, the new output is

Output remains 28 unchanged

But there must be a situation that cannot be divisible , stay PyTorch Can be seen in

Pooling is similar to

Shape:
        - Input: :math:`(N, C, H_{in}, W_{in})`
        - Output: :math:`(N, C, H_{out}, W_{out})`, where

          .. math::
              H_{out} = \left\lfloor\frac{H_{in} + 2 * \text{padding[0]} - \text{dilation[0]}
                    \times (\text{kernel\_size[0]} - 1) - 1}{\text{stride[0]}} + 1\right\rfloor

          .. math::
              W_{out} = \left\lfloor\frac{W_{in} + 2 * \text{padding[1]} - \text{dilation[1]}
                    \times (\text{kernel\_size[1]} - 1) - 1}{\text{stride[1]}} + 1\right\rfloor

stay PyTorch For conditions that cannot be divided, the default is to round down , You can choose to round up