When applying the convolution operation, one of the variations that can be applied to the process is to change the displacement units for the kernels. This parameter, which can be specified per dimension, is called stride. In the following image, we show a couple of examples of how stride is applied. In the third case, we see an incompatible stride because the kernel can't be applied to the last step. Depending on the library, this type of warning can be dismissed:
The other important fact when applying a kernel is that the bigger the kernel, the more units there are on the border of the image/matrix that won't receive an ...
