Deepstream or only tensorRT

Hi, Im using a camera.I want to use ssd model inference the image of the camera.I use the jetson GPU. And i alse performs image processing with opencv.I mean one camera and two processing methods.The results of the two methods are drawn on the image at the same time.
My question is whether to use deepstream or only tensorrt.
Thanks.

Hi,

It’s recommended to use Deepstream rather than OpenCV.

Deepstream optimizes the whole pipeline, memory usage and performance.
To use pure TensorRT, you will need to integrate OpenCV as a camera interface.

However, OpenCV doesn’t design or optimize for the shared memory system.
It use CPU memory as camera buffer which will lead to slow performance.

Thanks.

@AastaLLL
OK,thanks.
If I use the deepstream. How can I performs image processing with opencv in deepstream?
Where can I performs image processing with opencv in deepstream?
I should performs image processing with opencv after the ssd inference or the same time?
Are there any examples?

Hi,

You can check this sample for the deepstream -> OpenCV -> deepstream usecase:
https://devtalk.nvidia.com/default/topic/1047620/deepstream-sdk/how-to-create-opencv-gpumat-from-nvstream-/post/5397368/#5397368

Thanks.