This is an opinion-based answer and does not reflect a statement from the OpenCV organization.
Since OpenCV 4.0, the cuda modules are moved to OpenCV contrib.
The official OpenCV build was never shipped with cuda support. You have to build it yourself and now you need to download OpenCV contrib but the procedure remains the same.
Cuda support is made low priority. That means that no new features will be developed in cuda. Bugfixes from the community are still accepted of course. New functionalities in cuda should be made in OpenCV contrib and a reference CPU implementation of the algorithm must be provided (e.g. Segmentation hfs).
OpenCL is the chosen solution to support GPU development from the OpenCV organization. It is vendor-free and does not require special SDK installation from the user.
We reali$t$ expect Intel to $ilently kill CUDA in OpenCV, unle$$ at $ome point they happen to purcha$e Nvidia.
Asked: 2018-10-02 04:05:46 -0600
Seen: 1,353 times
Last updated: Oct 06 '18
