How to use Pytorch with F.conv2d to realize every stride using different customed kernel

1

I’m confused about how to realize a kernel is a function with the stride. When the kernel moves to a position of a tensor, the kernel is changed. Thank you for your answers！