How can I decode frame using GPU on Jetson Xavier

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

Hello everyone!

I am currently looking to read a frame (from a video), directly with the GPU using a Jetson Xavier.
The board is on Ubuntu 18.04 and Cuda 10.2

I already managed to read a video and upload each frame on the GPU with OpenCV C++, but now I would like to directly decode the frame using the GPU.
From what I understood I can’t directly read a frame with OpenCV (on GPU) because we need NvCuvid which is not available for Jetson Xavier.

So I know that VPI, VisionWorks and GStream are installed on my board. How can I use and which should I use to read a frame without using CPU.
Is there some code example in C++ ?

Thanks for your time and consideration.
Regards,
Paul

2

Moving to Jetson Xavier forum for resolution.

4

Hi,
In Xavier, the hardware video decoder is NVDEC. Please check the 00_video_decode sample:

/usr/src/jetson_multimedia_api/samples/00_video_decode/

https://docs.nvidia.com/jetson/l4t-multimedia/l4t_mm_00_video_decode_group.html

It demonsatrates decoding YUV frames into NvBuffer. And you can refer to cuda_postprocess() in 12_camera_v4l2_cuda to access the buffers through CUDA.