FPS with Three ZED X One Cameras

Software Python
, ,
1

Hi Stereolabs team,

I have a question regarding the expected frame rate when using three ZED X One cameras with a ZED Link Card and GMSL2 cables.
My hardware and software configuration is a Jetson Orin Nano Super Developer Kit running JetPack 6.2.1, ZED SDK 5.2.3, and stereolabs-zedlink-duo driver version 1.4.2.

Two of the cameras are for Virtual-stereo pair to calculate the distance to detected objects. The third camera operates independently and is used only for object detection, without depth computation

With this configuration, I am achieving approximately 8 FPS when using NEURAL depth mode. When switching to NEURAL_LIGHT, I see only a small improvement of around 1–2 FPS, with the total frame rate reaching approximately 10 FPS.
For object detection and segmentation, I am using yolov8s-seg.engine, .

Is this frame rate expected for this hardware and workload, or should the Jetson Orin Nano be able to achieve better performance?

Are there any recommended settings or optimizations that could improve the frame rate?

Thank you for your help!

2

Hi @NivC

These are not the expected frame rates with NEURAL and NEURAL_LIGHT depth modes.
Please find the expected values on this page of the documentation:

Are you using the Jetson in MAXN power mode with the jetson_clocks script running?
Are you running other software or algorithms that use the GPU?
Is your YOLO model optimized to run on the Jetson module?
Are you using YOLO with the ZED SDK as described here?