Running YOLOv8 on Jetson Nano with GPU Acceleration -JETSON Nano 16emmc

[fst19021]

Dear NVIDIA Support Team,

I am currently facing challenges running the YOLOv8 model on my Jetson Nano, as it is performing very slowly due to the model running on the CPU instead of utilizing the GPU. The issue seems to stem from the following:

1. The YOLOv8 model (via the Ultralytics package) requires Python 3.8, which I have installed in a separate virtual environment.
2. OpenCV with CUDA support and other dependencies compatible with Python 3.8 cannot be installed directly due to Jetson Nano’s Python 3.6 limitation in the default JetPack version.

To address this, I am considering upgrading to JetPack 5 to enable Python 3.8 compatibility and potentially resolve the GPU utilization issue. However, I am unsure if this is the best solution or if the Jetson Nano hardware and software limitations would still pose challenges.

My questions are:

1. Would upgrading to JetPack 5 allow me to use Python 3.8 and CUDA-supported libraries seamlessly on the Jetson Nano for YOLOv8?
2. If upgrading is not a viable solution, are there any recommended approaches to run YOLOv8 with GPU acceleration on the Jetson Nano while using Python 3.8?
3. Are there alternative methods or optimizations that could improve performance without compromising the environment setup for YOLOv8?

Thank you for your assistance. Any guidance or resources you could provide would be greatly appreciated.

[AastaLLL]

Hi,

JetPack 5 doesn’t support Jetson Nano.
To use a newer software release (ex. JetPack 5 or JetPack 6), please upgrade to our Orin series platform.

If this is not an option, please search for YOLOv8 on Jetson Nano.
(sorry that we cannot share third-party links here)

There are some tutorials that accelerate inference via TensorRT.
Thanks.