Try   HackMD

Table of Contents

YOLOv8 Tutorial

Installing Anaconda

  1. Download the installer from the official Anaconda website to your computer.

💡Tip

  • Please choose the appropriate installer based on your operating system.
  • All versions provide a graphical user interface for installation.
Windows Mac Linux
Download Download Download

Setting Environment Variables

  1. Search for Environment Variables in your system settings.

  2. Click on Environment Variables.

    image

  3. Under System Variables, select Path and click Edit.

    image

  4. Click New and add the following environment variable paths:

    image

Environment Variable Paths [Expand] 
C:\ProgramData\Anaconda3
C:\ProgramData\Anaconda3\Script\Library\bin
C:\ProgramData\Anaconda3\Scripts
C:\ProgramData\Anaconda3\condabin

Creating a YOLOv8 Environment

  1. Launch Anaconda Navigator.

    image

  2. At the bottom of the interface, click Create to add a new environment.

    image

  3. In the Create New Environment dialog, configure the environment as follows:

  • Name: YOLOv8
  • Location: C:\User\Ryan\envs\YOLOv8
    (Replace Ryan with your actual Windows username if different)
  • Packages: Python 3.10.16
    image

Activating the YOLOv8 Environment

  1. Select the newly created YOLOv8 environment and click Open Terminal.

    image

  2. You should now see the terminal window for the environment.

    image

Advanced Usage [Expand]

Steps 6~10 can be skipped by using the following commands:

  1. Open Windows Command Prompt (in Windows 11 it's renamed to Windows Terminal; we will refer to it as Windows Command here).
  2. Create the environment with the following command:
    conda create --name <environment_name> python=<python_version>
    Example:
conda create --name YOLOv8 python=3.10.16

image

  1. To activate the environment, use:
conda activate YOLOv8

image

  1. Once the environment is activated successfully, your terminal should look like this:
    image

Installing NVIDIA Driver

Installing the NVIDIA CUDA Toolkit

  1. Download and install CUDA Toolkit 11.8.
    image

Note

Be sure to select the correct version of your operating system.
This guide uses Windows 11 as an example.

  1. Go to the cuDNN Archive and download the version highlighted in the red box.
    image

Installing NVIDIA cuDNN

  1. Open the section titled Download cuDNN v8.9.7 (December 5th, 2023), for CUDA 11.x, and select the file as shown in the red box.

    image

  2. You must sign in to your NVIDIA Account to download.
    If you don’t have one, please register in a new tab.

    image

  3. After downloading, you will get the file:
    cudnn-windows-x86_64-8.9.7.29_cuda11-archive.zip

    image

  4. Locate the following path:
    C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8

    image

  5. Extract the contents of the ZIP file and copy the following folders into the CUDA directory above:

  • bin
  • include
  • lib

These should be merged into the corresponding folders inside C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8.

Verifying the NVIDIA Driver Installation

  • Open the Windows Command Prompt and type the following command to check if the GPU is detected by the driver:
nvidia-smi

If successful, you will see an output like the following:

image

  • Then, enter the following command to confirm the CUDA compiler (nvcc) is properly installed:
nvcc -V

image

Installing YOLOv8

  1. First, download the required files: 7-Zip, the YOLOv8 repository, and the pretrained YOLOv8n.pt model.

😁 Reminder
Please download the following files from the provided links before proceeding.

  • Download 7-Zip: Link
  • Clone or download the YOLOv8 repository: Link
  • Download the YOLOv8n.pt model: Link

  1. Install 7-Zip.
    To verify the installation, right-click on any file and check if the 7-Zip menu appears as shown below:

    image

  2. Extract the YOLOv8-main.zip file using 7-Zip or another archive tool.

    image

  3. After extraction, you may rename the folder to YOLOv8 and place it in a directory of your choice.
    For example:

    image

💡Tip
In this tutorial, we use D:\ as the example directory path.

  1. Move the yolov8n.pt model file into the YOLOv8 folder.

    image

  2. In the command line inside the YOLOv8 folder, run the following command to install the Ultralytics package:

pip install ultralytics

Installing PyTorch

  1. Visit the official PyTorch website to find the appropriate version for your environment.

  2. In the Windows Command Prompt, change the directory to your YOLOv8 project folder:

cd /d D:\YOLOv8

image
3. Install PyTorch using the following command:

🚨 Note
Replace pip3 with pip.

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

image
4. If this version does not work for your system, visit the Previous PyTorch Versionspage to find a suitable one.
image

Development Environment

Platform & Tools

Component Specification
Running environment System: Windows 11 Pro,
CPU: Intel Core i9-12900K,
RAM: 16GB * 2 (32GB),
GPU: NVIDIA GeForce RTX 3090
Visual Studio Code Sep 2024 (version 1.94)
Environment management Conda 24.9.0
Python Version 3.10.16

Package Requirements

Use the following command to install all necessary packages.
Ensure the environment matches exactly before using requirements.txt.

requirements.txt [Expand] 















































beautifulsoup4==4.12.3
certifi==2024.12.14
charset-normalizer==3.4.1
colorama==0.4.6
contourpy==1.3.1
cycler==0.12.1
filelock==3.13.1
fonttools==4.55.3
fsspec==2024.2.0
idna==3.10
Jinja2==3.1.3
kiwisolver==1.4.8
labelImg==1.8.6
lxml==5.3.0
MarkupSafe==2.1.5
matplotlib==3.10.0
mpmath==1.3.0
networkx==3.2.1
numpy==1.26.3
opencv-python==4.11.0.86
packaging==24.2
pandas==2.2.3
pillow==10.2.0
psutil==6.1.1
py-cpuinfo==9.0.0
pyparsing==3.2.1
PyQt5==5.15.11
PyQt5-Qt5==5.15.2
PyQt5_sip==12.16.1
python-dateutil==2.9.0.post0
pytz==2024.2
PyYAML==6.0.2
requests==2.32.3
scipy==1.15.1
seaborn==0.13.2
six==1.17.0
soupsieve==2.6
sympy==1.13.1
torch==2.5.1+cu118
torchaudio==2.5.1+cu118
torchvision==0.20.1+cu118
tqdm==4.67.1
typing_extensions==4.9.0
tzdata==2024.2
ultralytics==8.3.64
ultralytics-thop==2.0.14
urllib3==2.3.0

Dataset Preprocessing

Batch Video Converter

For converting videos recorded from various platforms (Windows, iOS, Android) into a unified MP4 format, use the Batch-Video-Converter tool.

  1. Unzip Batch-Video-Converter-main.zip and rename the folder to Batch-Video-Converter.

    image

  2. When prompted with Enter the input directory containing videos:, enter the path of the folder containing the videos to convert.

  3. When prompted with Enter the output directory for converted MP4 videos:, enter the desired output path.

    • If left blank, a folder named output will be created inside the input directory.

  4. When asked Do you want to keep audio in the converted videos? (Y/N, default: N):, input Y to retain audio, or N to remove it (default is N).

Video Frame Extractor

To simplify the labeling process for beginners, Video-Frame-Extractor helps convert videos into image sequences.

  1. Download and unzip the archive, then rename the folder to Video-Frame-Extractor.

    image

  2. Run the extractor:

python Video-Frame-Extractor.py

image
3. You'll be prompted to enter:

  • The input video directory
  • The output frame directory
  • The frame rate (e.g., 60 FPS is common)
  1. The extractor will process all videos in the folder and output numbered image frames starting from 0.
    image

Labelimg Annotation Tool

  1. Install the tool
pip install labelimg
  1. Download labelimg, unzip the archive, and rename the folder to labelimg.
    image
  2. Navigate into the folder:
cd labelimg

image
4. Launch the application:

python labelimg.py

image
5. The following window should appear:
image
6. Use Open Dir to label images in a folder or Open File for a single image.
image
7. After opening the folder, thumbnails appear in the bottom right, and the current image appears in the center.
image
8. In the top-left menu, enable Auto Save Mode under the View (V) tab.
image
9. Set the annotation mode to YOLO format.
image
10. Click Create RectBox to start labeling.
image
11. Use the cursor (displayed as a cross) to draw bounding boxes.
image
12. For custom labels like Helmet or Coverall, manually enter and confirm them.

🚨 Note
You must finish labeling all objects in one image before moving to the next. You cannot return to previous images.

image
image
13. All labeled items will appear in the box list on the right side.
image

📌 Pro Tip
- Path: D:\labelImg\data。

image
- File: predefined_classes.txt。
image
- Add label names to this file for quick label access during annotation.
image
image

  1. Once labeling is done, click Next Image.
    image
  2. In the terminal, you will see lines like:
Image:C:\Users\Ryan\Desktop\output\frame_0000.png -> Annotation:frame_0000.txt

This means frame_0000.png was labeled and saved as frame_0000.txt.

image
image
16. You are now in “Annotation Mode”! Already used labels will auto-complete from the menu.
image

🌐 Resources
For public datasets, consider RoboFlow 官網 and Kaggle 官網. Be sure to cite original authors.

Using YOLOv8

🔍 Basic Prediction Test

Run the following command to test the model with a sample image:

yolo predict model=yolov8m.pt source='https://ultralytics.com/images/bus.jpg'

​​​​:::spoiler Dataset Structure [Expand]
​​​​YOLOv8
​​​​└─dataset
​​​​│  ├─test
​​​​│  │  ├─images
​​​​│  │  │   ├─test_img001.jpg
​​​​│  │  │   ├─test_img002.jpeg
​​​​│  │  │   ├─test_img003.png
​​​​│  │  │   └─test_img004.tif
​​​​│  │  └─labels
​​​​│  │      ├─test_img001.txt
​​​​│  │      ├─test_img002.txt
​​​​│  │      ├─test_img003.txt
​​​​│  │      └─test_img004.txt
​​​​│  ├─train
​​​​│  │  ├─images
​​​​│  │  │   ├─train_img001.jpg
​​​​│  │  │   ├─train_img002.jpeg
​​​​│  │  │   ├─train_img003.png
​​​​│  │  │   └─train_img004.tif
​​​​│  │  └─labels
​​​​│  │      ├─train_img001.txt
​​​​│  │      ├─train_img002.txt
​​​​│  │      ├─train_img003.txt
​​​​│  │      └─train_img004.txt
​​​​│  └─valid
​​​​│     ├─images
​​​​│     │   ├─valid_img001.jpg
​​​​│     │   ├─valid_img002.jpeg
​​​​│     │   ├─valid_img003.png
​​​​│     │   └─valid_img004.tif
​​​​│     └─labels
​​​​│         ├─valid_img001.txt
​​​​│         ├─valid_img002.txt
​​​​│         ├─valid_img003.txt
​​​​│         └─valid_img004.txt
​​​​└─data.yaml
​​​​:::

Training YOLOv8 with Custom Dataset

1. Create the data.yaml File

Inside your YOLOv8 directory, create a file named data.yaml.

image
Example structure:

train: dataset/train
val: dataset/val
test: dataset/test

nc: 2
names: ['Helmet', 'Coverall']

📝 Parameter Descriptions

  • train: Path to the training set.
  • val: Path to the validation set.
  • test: Path to the test set.
  • nc : Number of classes.
  • names: Class names in order.0 corresponds to Helmet, 1 to Coverall.

2. Check Your Label Format

A label file (frame_0000.txt) may look like this:

15 0.402734 0.114583 0.099219 0.159722
16 0.417969 0.538889 0.178125 0.744444

📝 Explanation

  • The first value is the class ID (e.g., 16, 15).
  • The next four values represent normalized bounding box coordinates:
    <x_center> <y_center> <width> <height>
    If your label class IDs are not 0 and 1, you can use the YOLO-Class-ID-Replacer to remap them:
# Class ID mapping example
class_id_mapping = {
    "15": "0",  # Map class 15 to 0
    "16": "1",  # Map class 16 to 1
}

3. Start Training

Use the following command to begin training:

yolo detect train data=data.yaml model=yolov8n.pt epochs=100 imgsz=640 conf=0 device=0

Training Logs
YOLOv8 will:

  • Scan your train, val, and test folders
  • Display GPU and dataset configuration
  • Start training from 1/100 to 100/100 epochs

📝 Training Metrics

  • Epoch: Current training round.
  • GPU mem: GPU memory usage.
  • box loss: Bounding box regression loss.
  • cls loss: Classification loss.
  • dfl loss: Distribution focal loss (YOLOv8-specific).
  • Instances: Number of objects processed in current batch.

📝 Validation Metrics

  • Size: Input image size.
  • Class: Number of detected classes.
  • Images: Number of images in validation batch.
  • Box (P): Precision score.
  • R: Recall score.
  • mAP50: Mean Average Precision at IoU 0.50.
  • mAP50-95: Average precision across IoU thresholds from 0.50 to 0.95.

Model Checkpoints

After training, two model files will be generated:

💭 Model Descriptions

  • best.pt:
    The model checkpoint with the highest performance on the validation set.
    Use this for final prediction or deployment.
  • last.pt:
    The checkpoint from the final epoch. Useful for further fine-tuning, not necessarily the best performing.

4. Run Inference with Custom Model

Use the trained model to make predictions:

yolo detect predict model=best.pt epochs=100 imgsz=640 conf=0 device=0 source=0 save=True show=True

📝 Inference Parameters Explained

  1. model: Path to model checkpoint
    • Use pre-trained (yolov8n.pt) or your custom model (best.pt or last.pt)
  2. epochs: Epoch count for further training (ignored in prediction mode)
  3. imgsz: Image resolution (e.g., 640, 1280)
  4. conf: Confidence threshold
    • conf=0.25 will filter out predictions below 25% confidence.
  5. device: Device to use for prediction
    • 0 = first GPU, cpu = CPU, mps = Apple Silicon GPU
  6. source: Input source
    • 0 = webcam
    • images/sample.jpg = single image
    • videos/sample.mp4 = video file
    • Folder path or livestream URL also accepted.
  7. save: Whether to save output images/videos
    • True = save output to /runs/
    • False = do not save
  8. show: Display results in a pop-up window
    • True = show prediction
    • False = headless mode

Development Log

This YOLOv8 tutorial and deployment guide was authored and refined by Yu-Cheng Chang (張育丞), as part of a structured effort to document the full setup, training, and inference process of an object detection system using Ultralytics YOLOv8.

The motivation behind this guide stems from real-world needs in dataset preprocessing, efficient model training, and reproducible development pipelines. The tutorial was developed in an engineering-research context, especially focusing on:

  • ⚙️ Cross-platform setup (Windows 11)
  • 🐍 Environment isolation using Conda
  • 🧠 Deep learning framework installation including CUDA, cuDNN, and PyTorch (GPU acceleration enabled)
  • 📦 Ultralytics YOLOv8 usage, from simple prediction to full dataset training
  • 📽️ Dataset tools:
    • Batch-Video-Converter for MOV/WMV to MP4 conversion
    • Video-Frame-Extractor for extracting frames from video at a desired FPS
    • labelImg with detailed annotation instructions for YOLO format
    • 🧾 Model class remapping using YOLO-Class-ID-Replacer
      The entire tutorial has been written in both Chinese (for HackMD and local use) and translated into clean, professional English Markdown format for GitHub sharing and open-source collaboration.
      The teaching structure was organized progressively to ensure that students, researchers, and developers could:
  1. Follow step-by-step instructions without missing key environment issues.
  2. Understand practical training parameters and their meanings.
  3. Deploy and test models with confidence on real-time data.

This document serves not only as technical guidance but also as a reproducible lab notebook — one that balances accessibility and precision.

Update History

Version Date Description
1.0 Jan 01, 2025 Initial draft created
2.0 Jan 15, 2025 Added CUDA and cuDNN setup instructions
3.0 Jan 20, 2025 Integrated video conversion pipeline
3.1 Jan 22, 2025 Updated video frame extraction documentation
4.0 Jan 26, 2025 Added labeling tool (labelImg) tutorial
5.0 Feb 03, 2025 Integrated YOLOv8 training guide
5.1 Feb 17, 2025 Added YOLO-Class-ID-Replacer explanation
6.0 Mar 07, 2025 Organized complete data.yaml usage details
6.1 Mar 22, 2025 Revised training metrics and evaluation logs
7.0 Apr 16, 2025 Added inference demo and explanation
8.0 May 04, 2025 Finalized English version for GitHub release

Citation

@article{chang2025yolov8tutorial,
  title     = {YOLOv8 Tutorial},
  author    = {Yu-Cheng Chang},
  journal   = {HackMD},
  year      = {2025},
  url       = {https://hackmd.io/@Yucheng208/YOLOv8-Tutorial}
}

License

This content is licensed under the MIT License. See the LICENSE for details.

MIT License

Copyright (c) 2025 Yu-Cheng, Chang

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Contributing

Contributions or feedback are welcome!
Please feel free to contact me via email for any suggestions or inquiries.

Last changed by 
 
Yu-Cheng Chang 張育丞
Dept. of Computer Science and Information Engineering, National Taipei University of Technology, Taipei City, Taiwan (R.O.C)
0
488

Read more

Cloud and IoT Security Exam

()[]

Apr 14, 2025
Linear Algebra 線性代數

Linear Algebra Assignment Grading Guidelines 線性代數作業批改規定

Feb 19, 2025
Read more from Yu-Cheng Chang 張育丞
Published on HackMD