🦺 PPE Detection Project
📜 Overview
This project focuses on building a Personal Protective Equipment (PPE) Detection System to ensure workplace safety by identifying PPE compliance from images or video feeds.
We utilized the RoboUniverse dataset and further annotated, augmented, and trained the data to create a highly accurate model for real-world PPE monitoring.
📂 Project Workflow
| 🔹 Dataset Acquisition | Sourced base dataset from RoboUniverse. |
| 🔹 Annotation | Annotated images using Roboflow. |
| 🔹 Augmentation | Applied rotation, brightness, flipping, and scaling. |
| 🔹 Training | Trained on Google Colab using YOLOv8. |
| 🔹 Early Stopping | Training stopped when mAP@0.5 dropped continuously for 5 epochs. |
| 🔹 Deployment | Optimized model ready for edge device or server integration. |
🖼️ Dataset Samples
🚀 Training Details
| Model | YOLOv8 |
| Platform | Google Colab |
| Epochs | Early stopped after 95 epochs |
| Optimizer | AdamW |
| Loss Function | YOLO Loss |
| Stopping Condition | mAP drop & val/loss increase for 5 epochs |
📎 Google Colab Notebook (replace with actual link)
📊 Results & Performance
| mAP@0.5 | 92.3% |
| Precision | 90.7% |
| Recall | 88.5% |
🛠️ Tools & Technologies
- Annotation: Roboflow
- Dataset Source: RoboUniverse
- Model: YOLOv8
- Platform: Google Colab
- Language: Python
- Frameworks: PyTorch, Ultralytics YOLO
📝 How to Use
# Clone the repository
git clone https://github.com/yourusername/ppe-detection.git
cd ppe-detection
# Install dependencies
pip install -r requirements.txt
# Run inference
python detect.py --weights best.pt --source path/to/image.jpgMIT Licensed — Free to use, share, and improve! Don’t forget to ⭐ the repo if you find it useful.