Can YOLOv8-Seg inference run in near real-time on Jetson Nano?

platform

Hello everyone,
I'm working on a project that requires YOLOv8-Seg for image segmentation, and I’m aiming for inference speeds as close to real-time as possible. I’m currently using a Jetson Nano with JetPack 4.6.1 installed.

I’ve been trying different approaches to optimize performance, but I’m not sure if near real-time speeds are achievable on this setup. If anyone has experience with YOLOv8-Seg or similar segmentation models on the Jetson Nano, I’d appreciate any advice or recommendations on achieving faster inference.

Thank you in advance for your help!

jakaskerl

platform
Where is the bottleneck? How fast are the frames streaming from the OAK device?

Thanks,
Jaka

platform

jakaskerl
It is 60fps.

jakaskerl

platform
And you wish to achieve more? Which device are you using? Generally IMX378 can not stream more than 60FPS. With NN model running on top, this is probably lower.

Thanks,
Jaka

platform

What kind of program code should be executed to achieve this?

jakaskerl

platform
https://docs.luxonis.com/software/depthai/examples/yolov8_nano/ or the cpp counterpart.

Thanks,
Jaka