Face Mask Detection Using YOLOv8 😷

A real-time face mask detection system using Ultralytics' YOLOv8 object detection models. This project detects whether a person is wearing a mask or not, leveraging annotated datasets and custom-trained models.

---

🌐 Project Overview

This repository contains all the code and assets to:

• Convert XML annotations to YOLO format
• Train YOLOv8 (both yolov8n and yolov8s models)
• Run real-time mask detection via webcam
• Visualize training logs with TensorBoard

---

📁 Project Structure

face-mask-detection/
├── data/                        # Contains dataset and annotation scripts
│   ├── images            
│   └── labels                   # XML annotations
├── dataset.yaml                 # YOLO dataset configuration
├── notebook/
│   └── your_notebook.ipynb      # Training and experimentation notebook
├── runs/                        # YOLOv8 training output (weights, logs, plots)
│   ├── detect/                  #runs/train/face_mask_detector contain logs for YOLOv8n
│   ├── train/                   #runs/s_runs/s_train/face_mask_detector_s contaons logs for YOLOv8s
│   └── ...
├── yolov8n.pt                   # YOLOv8n pre-trained model (used initially)
├── yolov8s.pt                   # YOLOv8s pre-trained model (improved results)
├── realtime_mask_detection.py   # Script to run real-time detection
├── log_to_tensorboard.py        # TensorBoard logging (optional)
├── xml_to_yolo.py               # Converts Pascal VOC XML to YOLO format
├── test1.jpg, test2.jpg         # Sample images for testing
└── README.md                    # This file

---

🔧 Installation

1. Clone the Repository

git clone https://github.com/AksharKher-30/Real-time-face-mask-detection.git
cd face-mask-detection

2. Install Dependencies

pip install -r requirements.txt

Make sure you have Python 3.8+ and pip installed on your system.

---

📁 Data Preparation

• Annotation are provided in Pascal VOC (XML) format.
• Convert to YOLO format using the following command:

python xml_to_yolo.py

This script will:

• Parse all .xml files
• Convert annotations to YOLO format
• Save converted .txt files in the corresponding labels/ directories

Dataset format expected by YOLO:

data/
├── test/
│   ├── annotation/
│   ├── images/
│   └── labels/
├── train/
│   ├── annotation/
│   ├── images/
│   └── labesl/

Update dataset.yaml accordingly with the correct paths and class names:

train: data/images/train
val: data/images/test
nc: 2
names: ['Mask', 'No Mask']

The trained weights will be saved under:

runs/s_runs/s_train/face_mask_detector_s/weights/
├── best.pt
└── last.pt

best.pt is used for inference.

---

🎥 Real-Time Detection

To run live webcam detection using the trained model:

python realtime_mask_detection.py

Make sure to set the correct path to your trained model weights, e.g. best.pt, inside the script.

---

TensorBoard Logging (Optional)

To visualize training metrics with TensorBoard:

tensorboard --logdir runs/

Then open http://localhost:6006 in your browser.

---

📊 Performance

📉 Live Face Mask Detection Accuracy

Model	Mask Accuracy	No Mask Accuracy
YOLOv8n	66.2%	49.5%
YOLOv8s	84.5%	81.9%

✅ YOLOv8s clearly outperformed YOLOv8n in real-time detection.

---

📊 Results Example

Here is a snapshot of the real-time detection in action:

This is for YOLOv8s pre-trained model