Does pytorch support convolution with different stride in different direction?

1

For example, the convolution kernel size is 333, and the stride is 122,

2

Yes, this should work. You would have to pass these values as tuples to the arguments.

3

thanks, I just read the documentation and find it works