For https://www.intel.ai/winograd-2/ , why use stride = 2 ?
Why need to transform input image pixels ?
Why this C++ implementation of winograd convolution does not require any input tensors transformation ?
$\begingroup$
$\endgroup$
Add a comment
|