Calling an Opencv function inside a Cuda Kernel

Question

I intend to perform a panorama of 4 images. For this, I intend to process images 1 and 2, 2 and 3 and then 3 and 4 and get the homographies and then stitch the images.

Now, getting the SIFT features in each image can be parallelized. So, using CUDA on a GPU, I intend to call Surffeatureextractor and descriptor in each kernel after having passed the image to the kernel.

Now, I wish to know if it is possible to write an opencv function inside a cuda kernel? for example, if my kernel is

    extract_sift(image, *keypoints, *descriptor)
    {
    //I want to call sift descriptor from an external opencv lib here }

Please let me know how to do that or direct me to certain links that help me do that.

OpenCV has a gpu library now if you want to give it a try. opencv.willowgarage.com/wiki/OpenCV_GPU. But you still can not call it from inside a kernel. — Pavan Yalamanchili
– Pavan Yalamanchili, Commented Apr 17, 2012 at 14:41

eudoxos · Accepted Answer · 2012-04-17 06:32:32Z

2

What you ask is impossible. If you understand the OpenCL architecture, it is clear why: kernels run on a different device, have different memory, architecture etc, and all communication with them is via buffers. You can pass your descriptor as a struct (pointer to that struct, more precisely) to the kernel, provided you also declare that same struct in the OpenCL code.

answered Apr 17, 2012 at 6:32

eudoxos

19.2k11 gold badges71 silver badges129 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

Calling an Opencv function inside a Cuda Kernel

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related