How to transfer data to cuda-process of nvivafilter

Hi,
Recently I am researching how to use cuda-process of nvivafilter.I downlaod an nvsample_cudaprocess source code and modify the contents of gpu_process. I used gstreamer to write a program, and I need to transfer some parameters to the gpu_process of nvivafilter.but I don’t know how to send it. Can someone help me?

Hi changding, by following README you can build your own libnvsample_cudaprocess.so. Any difficulty you have faced in modifying nvsample_cudaprocess.cu?

Hi DaneLLL
What should I do if I want to transfer external variables to the gpu_process interface of the nvivafilter element?

hi changding, you can extern the variables to libnvsample_cudaprocess.so

Hi DaneLLL
The input image data is stored in EGLImageKHR, and the image data output to the next element is also this EGLImageKHR right?

Some examples of using nvivafilter:
https://devtalk.nvidia.com/default/topic/963123/jetson-tx1/video-mapping-on-jetson-tx1/post/4979740/#4979740
https://devtalk.nvidia.com/default/topic/978438/jetson-tx1/optimizing-access-to-image-data-acquired-with-nvcamerasrc/post/5026998/#5026998

Hi changding,

You can also review GstCuda and check if it matches your requirements:

http://developer.ridgerun.com/wiki/index.php?title=GstCUDA

  • Carlos R.

Hi DaneLLL:
I also have the same question, but cannot find any good answer from your links. I have some cuda algorithms, which can not modify the input EGLImageKHR image in place but create new image. I know I can copy the new generated image to the input, but is there any better way like swap the new image handle to it or other? In short, I want to create a gstreamer transform element. Thanks!

1 Like

Hi cnhzcy14,
We suggest you use MMAPI to have more flexibility.