How to access to image array in GPU buffer without copy to CPU buffer?

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

• Hardware Platform (Jetson / GPU):Jetson
• DeepStream Version:5.0
• JetPack Version (valid for Jetson only):4.4
• TensorRT Version:7.x

Hi,
I want to access image array from gstreamer buffer and the feed to the custom model, and I see this example,
But one big problem of this example is that copying frame from GPU buffer to CPU buffer, But I don’t want to copy to CPU buffer, because I have to convert to Tensor on CUDA and then feed to model.
I want to have image array in GPU buffer without copy to CPU buffer.

I write custom gstreamer plugin with chain function, and do custom inference model in that plugin.

In that example for copying image array into CPU buffer is used numpy lib, I want to know is it possible to use CuPy lib or other ML framework to access to image array in GPU memory?

3

Sorry to tell that we currently don’t have bindings for GPU buffer. It is on our roadmap for future release.

Thanks