How to access the NvBufSurface GPU pointer in Python?

[polysong]

• Hardware Platform (Jetson / GPU) : dGPU
• DeepStream Version: 7.0

Hi,

I am configuring a DeepStream pipeline in Python to process multiple video stream.

I added a probe on the src pad of nvosdretrieve the frame. but I have a couple of challenges

I would like to keep the buffer in NV12 and map the frames directly to GPU memory so they can be rendered to the screen using OpenGL.

I was reviewing the deepstream python api documentation, and see the following:

pyds.get_nvds_buf_surface maps the buffer to CPU and only supports RGBA

pyds.get_nvds_buf_surface_gpu return a pointer to the GPU buffer, but it only supports X86 and RGBA format.

I found the following information in other forum posts:

From my own research I figured out that surface->surfaceList[i].dataPtr can be used on dGPU, while on Tegra you have to chain NvBufSurfaceMapEglImage and cuGraphicsResourceGetMappedEglFrame like described here.

My question is how to retrieve an NvBufSurface in a pad probe function using DeepStream Python, and then access the surfaceList as well as the corresponding GPU pointer from it. Are there any relevant examples or function available?

Thank you!

[fanzh]

Regarding “be rendered to the screen using OpenGL.”, why not use DeepStream rendering plugin?
Curretly, In python code, only one function of accessing NvBufSurface GPU memory is get_nvds_buf_surface_gpu, but it only supports RGBA format, as commenting in the code.

[polysong]

Thank you for the clarification. This resolves my question.

I am building a PyQt application that uses DeepStream for video streaming and renders the frame in QOpenGLWidget (Qt Widget)，which requires access to the NvBufSurface.