Use of pytorch (to be supplemented)

nn.Conv2d(3, 48, kernel_size=11, stride=4, padding=2)
Represents the entered channel by 3, Output channel by 48. Convolution kernels for 11x11.
among padding for int or tuple.padding=2： Fill two circles of zeros around ,padding=(1,2)： Means fill in a line of zero at the top and bottom , Two rows of zeros shall be added on the left and right sides respectively . And you can use nn.ZeroPad2d() Zero compensation （ own google）