There is a variable called CUDA_TOOLKIT_DIR in the cmake configuration.
This variable is initialized by
cmake with the location of the CUDA toolkit, by default
However, this settings yields error messages in the link step, claiming that
ld can’t use the CUDA directory.
The thing is, the content of this CUDA_TOOLKIT_VARIABLE variable seems to passed directly to
Thus, it needs to point to the actual location where
ld will find the libraries to link OpenCV with, i.e.,
/Developer/NVIDIA/CUDA-6.x/lib (note the extra lib at the end).
Wow, that was easy, but took me a bit of time to figure it out. Hope this will help someone else!