Using NVIDIA Argus with YUV422 GMSL2 Camera on Jetson Orin Nano (JetPack 6.0 / L4T 36.3)

Robotics & Edge Computing Robotics - Isaac Isaac ROS
,
1

Hello,

We are working on a NVIDIA Jetson Orin Nano (8 GB) running JetPack 6.0 / L4T 36.3 (kernel 5.15.136-rt-tegra) on a custom carrier board.

The system uses a GMSL2 monocular camera that outputs YUV422 @ 8 bit (UYVY8_1X16).

We’d like to know if it’s possible to use NVIDIA Argus with this camera, and if so, how.

As far as we understand, Argus typically requires RAW Bayer sensor data to enable ISP-based processing and GPU image pipelines, but we would like to use Argus or a similar GPU-accelerated path for processing frames from this YUV GMSL2 camera, rather than relying solely on CPU-side V4L2 or GStreamer pipelines.

Specifically:

  • Can Argus be configured to accept YUV422 input instead of Bayer RAW?

  • If not, is there any supported workaround (for example, using nvarguscamerasrc through an intermediate)?

  • Alternatively, is there a recommended way to push V4L2 YUV frames directly into GPU memory (e.g., CUDA) to achieve equivalent performance?

  • If it’s not currently supported, is there anything in the NVIDIA roadmap to support YUV422 in the future?

Media infos:

media.log (1.8 KB)

Thanks in advance.

2

You can simply use nvv4l2camerasrc instead of nvarguscamerasrcfor GStreamer pipelines.

3

Hello Patrick,

Thanks for the reply.

Is the nvv4l2 actually using the ISP?

Our ultimate goal would be to let the ISP handle the image processing (as Argus does with Bayer Raw data).