README.md

Ultralytics YOLO11n + ByteTrack (Object Tracking) Optimization test

This is a repository of test codes for YOLO11n object detection and tracking code.
The code optimize is done only for object detection part. ByteTrack algorithm are used as is without any optimization.

File Name	Using Ultralytics framework	Device	Description	FPS(reference)
yolo11n_ultralytics_cpu.py	✓	CPU	Run original Ultralytics YOLO11n with CPU	4
yolo11n_ultralytics_openvino.py	✓	CPU/iGPU/NPU(/dGPU?)	Used OpenVINO to optimize. Still using Ultralytics framework	CPU 13 iGPU 30
yolo11_ultralytics_cuda.py	✓	NV GPU	Used TensorRT to optimize. Still using Ultralytics framework	60
yolo11_ov_bytetracker.py		CPU/iGPU/NPU/dGPU	Use OpenVINO (without Ultralytics framework) to run object detection, and pass the result to ByteTrack to track the detected objects	CPU 28 iGPU 60 dGPU 72
yolo11_hailo_bytetrack.py		Hailo-8	Use Hailo-8 and HailoRT for object detection without Ultralytics framework	120

• Benchmark HW1: Note PC, Dynabook SZ/LVL W6SZLV5FAL, Core i5-1235U + Hailo-8

• Benchmark HW2: Desktop PC, Core i7-10700K + RTX-3050 / Arc A380

• OS: Windows 11

• OpenVINO: 2025.2.0

How to setup the environment

1. Install required Python modules

Windows

python -m venv venv
venv\Scripts\activate
python -m pip install -U pip
pip install -U setuptools wheel
pip install -r requirements.txt

Linux

python3 -m venv venv
. venv/bin/activate
python3 -m pip install -U pip
pip install -U setuptools wheel
pip install -r requirements.txt

Note:

• You need to install TensorRT for CUDA/TRT version, and HailoRT for Hailo-8 version. Please go and refer to the vendor's web documents for the details.

• You need to download yolo11 model from Hailo Model Zoo ('yolov11n.hef')

Demo images

]

Movie

Remarks

• Hailo-8 is installed in a m.2-Thunderbolt4 box, and connected to the notebook PC via USB-C cable (Thunderbolt) in my test