Installing and Setting Up Ultralytics YOLOv8 on Jetson Orin Nano with JetPack SDK 6.0

Issue Overview

Users are seeking guidance on installing and configuring Ultralytics YOLOv8 (version 8.2.48) on the Nvidia Jetson Orin Nano development board running JetPack SDK 6.0. The specific requirements include:

• Using GPU acceleration for YOLO object detection
• Installing Ultralytics with CUDA and cuDNN support
• Utilizing PyTorch 2.3.1 and torchvision 0.18.1
• Verifying GPU utilization for the YOLO model

This setup is crucial for leveraging the Jetson Orin Nano’s AI capabilities and ensuring optimal performance for object detection tasks.

Possible Causes

While the forum discussion doesn’t explicitly mention issues, potential challenges in this setup could include:

1. Compatibility conflicts: Mismatches between Ultralytics, PyTorch, and JetPack SDK versions may cause installation or runtime errors.
2. Incorrect CUDA configuration: Improper CUDA setup can prevent GPU acceleration, leading to suboptimal performance.
3. Missing dependencies: Absence of required libraries or drivers may hinder the installation or execution of Ultralytics YOLOv8.
4. Limited system resources: Insufficient memory or storage on the Jetson Orin Nano could impede installation or model execution.
5. Outdated software: Using older versions of JetPack SDK or other components may lead to compatibility issues with the latest Ultralytics release.

Troubleshooting Steps, Solutions & Fixes

To install and set up Ultralytics YOLOv8 on the Jetson Orin Nano with JetPack SDK 6.0, follow these steps:

1. Prepare the Jetson Orin Nano:

   • Ensure JetPack SDK 6.0 is properly installed on your Jetson Orin Nano.
   • Update the system:

     sudo apt update && sudo apt upgrade -y
     

2. Install dependencies:

   sudo apt install python3-pip libopenblas-base libopenmpi-dev
   

3. Install PyTorch and torchvision:

   • Follow the official NVIDIA instructions for installing PyTorch 2.3.1 and torchvision 0.18.1 on Jetson platforms.

4. Install Ultralytics:

   pip3 install ultralytics==8.2.48
   

5. Verify CUDA and cuDNN support:

   import torch
   print(f"CUDA available: {torch.cuda.is_available()}")
   print(f"cuDNN version: {torch.backends.cudnn.version()}")
   

6. Run a sample YOLOv8 inference:

   from ultralytics import YOLO
   
   # Load a pretrained YOLOv8n model
   model = YOLO('yolov8n.pt')
   
   # Run inference on an image
   results = model('path/to/image.jpg')
   
   # Display results
   results.show()
   

7. Check GPU utilization:

   • During inference, open a new terminal and run:

     tegrastats
     

   • Look for GPU usage percentage in the output.

8. Optimize performance:

   • Use TensorRT for faster inference:

     model = YOLO('yolov8n.pt')
     model.export(format='engine')  # Export to TensorRT
     

   • Run inference with the optimized model:

     model = YOLO('yolov8n.engine')
     results = model('path/to/image.jpg')
     

9. Troubleshooting:

   • If you encounter "CUDA out of memory" errors, try reducing the input image size or batch size.
   • For any installation issues, consult the official Ultralytics documentation for Jetson devices.

By following these steps, you should be able to successfully install and run Ultralytics YOLOv8 on your Jetson Orin Nano with GPU acceleration. Remember to monitor system resources and adjust model parameters as needed for optimal performance on the Jetson platform.