Conv2d

class torch.nn.Conv2d(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, bias=True)

 

padding x 2 + stride == kernel_size时，步长是几则缩小几倍