YOLOv8Lab

Welcome to the YOLOv8Lab repository, a dedicated platform for setting up and using YOLOv8 for object detection tasks in educational settings.

This repository and code is based on the YOLOv8 project by: Jocher, G., Chaurasia, A., & Qiu, J. (2023). Ultralytics YOLO (Version 8.0.0) [Computer software]. https://github.com/ultralytics/ultralytics

Table of Contents

1. Installation
2. Labeling
3. Training
4. Predictions
5. Tracking

Installation

This project assumes that Anaconda and Spyder are installed on your Windows system. Follow the steps below to set up your environment.

Prerequisites

• Anaconda: Download and install Anaconda from the official website. See installation guides for different operating systems:
  • Installing on Windows,
  • Installing on Linux,
  • Installing on macOS.
• Python >= 3.9 (comes with Anaconda)

Directory structure

1. Create the yolov8 project folder

C:/
├─ Users
│   ├─ UserName
│   │   ├─ yolov8                  
│   │   │   ├─ labeling/        
│   │   │   ├─ train/
│   │   │   ├─ test/     
│   │   │   └─ valid/
│   │   └─ ...             
│   └─ ...
└──...

Environment Setup (via Anaconda Prompt)

2. Create a new conda environment (in the windows search bar type "Anaconda Prompt"):

   conda create --name yolov8 python=3.9 spyder=5

3. Activate the environment:

   conda activate yolov8

4. Install Required Packages:

   pip install ultralytics

5. Launch Spyder within the new environment:

   spyder

6. … and check your installation by opening and running the yolov8_test.py in Spyder:

   Download yolov8_test.py

7. ... or by typing the following commads in your Command Line Interface (CLI):

   Predict on an image

   yolo predict model=yolov8n.pt source='https://ultralytics.com/images/bus.jpg'

   Run a pretrained yolo object detector on your webcam

   yolo predict model=yolov8n.pt source=0 show=True

Labeling

Additional Tools

1. Install labelImg for image annotation using the same activated yolov8 environment:

   conda install -c conda-forge labelimg

2. Start the labeling software:

   labelimg 

3. ... in the correct folder (labelImg [path to images] [path to predefined_classes.txt file]):

   labelimg C:\Users\UserName\yolov8\labeling C:\Users\UserName\yolov8\labeling\predefined_classes.txt 

   

Training

Start a Training on your local machine directly in the CLI ...

yolo task=detect mode=train model=yolov8n.pt imgsz=800 data=path/to/boars.yaml epochs=200 batch=8 project=/path/to/your/project/training_runs/ name=yolov8n_imgsz800 device="cpu"

... or using a python environment

from ultralytics import YOLO
import os

# change the working directory so that the training results are saved to the correct folder.
os.chdir("path/to/yolov8/dataset/training_runs/")
# Load a pretrained model
model = YOLO("yolov8n.pt")

# train the model (transfer learning)
model.train(data="path/to/yolov8/boars.yaml",
            epochs=20,
            imgsz=800,
            batch=8,
            project = "path/to/yolov8/training_runs/",
            name="yolov8n_imgsz800",
            device="cpu")  # 0 for GPU (check pytorch installation hints) or "cpu"

• model=yolov8n.pt: Indicates the model to be used. Here, the YOLOv8 Nano model (yolov8n.pt) is used.
• imgsz=800: Determines the size of the input images in pixels. In this case, the image size is 800x800 pixels (default: 640).
• data=path/to/boars.yaml: Specifies the path to the data file that defines the training and validation data.
• epochs=200: Sets the number of training epochs. Here, it is 200 epochs.
• batch=8: Determines the batch size, i.e., the number of images processed simultaneously. Here, it is 8 (default 16).
• project=/path/to/your/project/training_runs/: Indicates the project directory where the training runs will be saved.
• name=yolov8n_imgsz800: Sets the name of the training run. This helps distinguish between different runs.
• device="cpu": Specifies the CPU to be used for training. Alternatively, GPUs 0 and 1 can be used (device=(0,1)).

Check here for additional train settings and hyperparameters

Example boars.yaml file

# Contents inside the .yaml file

train: path\to\yolov8\train
val:   path\to\yolov8\valid
test:  path\to\yolov8\test

# total number of classes
nc: 2
names: ['boar','human']

Predictions

CLI

yolo predict model= .\training_runs\yolov8n_imgsz800\weights\best.pt source=.\test project=predictions name=yolov8n_imgsz800 conf=0.2 imgsz=800

Python environment

from ultralytics import YOLO
import cv2
from matplotlib import pylab as plt

# Load a trained model
model = YOLO("path/to/yolov8/training_runs/yolov8m_imgsz800/weights/best.pt")

# Use the model (predict on an image)    
results = model("path/to/yolov8/test",
                project = "path/to/yolov8/predictions",
                name = "yolov8m_imgsz800",
                show=True, 
                show_labels=False, 
                conf=0.2,
                imgsz=800,
                classes=None,
                save=True
                ) 

#%%     
# Plot the results (within matplotlib)
res_plotted = results[0].plot() # results[0]: show the first image
res_plotted_rgb=cv2.cvtColor(res_plotted, cv2.COLOR_BGR2RGB) 
plt.imshow(res_plotted_rgb)

Predictions (quantitative evaluation on independent images)

The images must, of course, be labeled but should not have been used in training or validation.

yolo detect val model=path\to\weights\best.pt data=path\to\yourFile.yaml

Tracking

yolo track model=path\to\weights\best.pt project=path\to\trackings\ name=yolov8n_800_botsort source="path\to\test\" tracker=botsort.yaml

Here you can find the .yaml files for your tracker, as well as an additional script if you want to draw lines for your tracked paths.