Vietnam Traffic Sign Detection

Our first Machine Learning, Deep Learning, and Computer Vision Project

---

Contributors:

• 🧠 David Vu
• 💻 The Peak
• 💻 Hoàn Bự Con

---

About the Project:

• Dataset: Vietnam Traffic Signs

• The model is based on YOLOv8-medium architecture

• This project is designed with strong potential for future development and improvement.

• This shit is so peak

---

Languages and Tools:

---

Installation guide:

📁 Project Structure

Traffic-Sign-Detection/
│
├── src/
│   ├── core/
│   │   └── utils.py
│   │
│   ├── training/
│   │   ├── train.py
│   │   └── config.py
│   │
│   ├── weight/
│   │   └── all_weight/
│   │       ├── trainX
│   │       │   ├── best.pt
│   │       │   └── last.pt
│   │       │
│   │       └── yolov8m.pt
│   │
│   ├── pipeline/
│   │   ├── predict.py
│   │   ├── real_time_predict.py
│   │   ├── real_time_predict_smooth_advanced.py
│   │   └── real_time_predict_nlp_hybrid.py
│   │
│   └── sort.py
│
├── data/
│   ├── dataset/
│   ├── input/
│   ├── output/
│   └── real_time_output/
│
├── assets/
├── runs/
├── .env
├── requirements.txt
└── README.md

1️⃣ Clone the repository

git clone https://github.com/HoanBuCon/Traffic-Sign-Detection
cd Traffic-Sign-Detection

2️⃣ (Optional) Create a virtual environment

python -m venv venv
# Activate virtual environment
source venv/bin/activate       # Linux/macOS
venv\Scripts\activate          # Windows

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Prepare dataset and config

• Dataset layout (YOLO format):
  • Images: dataset/images/train, dataset/images/val, dataset/images/test
  • Labels: dataset/labels/train, dataset/labels/val, dataset/labels/test
• If your images/labels are flat under dataset/, split into train/val:

  # From repo root
  python -m src.split_dataset

• Dataset YAML (auto-detected):
  • Preferred: data/dataset/data.yaml (current setup)
  • Fallbacks: data/data.yaml, then repo-root data.yaml
  • Override via env var:

  $env:DATA_YAML = "data/dataset/data.yaml"

• Folders are auto-created at runtime; no manual setup needed:
  • input/ (for your images), output/ (predictions under output/predictX/{images,json}), real_time_output/ (saved videos), all_weight/ (trained weights organized as trainX/).
• Place any test images you want to run in input/.

5️⃣ (Optional) Use Pre-trained Weights to Run the Project

• Instead of retraining the model from scratch, you can use our pre-trained YOLOv8 weights trained on a 12-class traffic sign dataset available here:
• 👉 Download Pre-trained Weight (Google Drive)

⚠️ After downloading, place the weight files in the following directory structure .\Traffic-Sign-Detection\weight\all_weight:

Traffic-Sign-Detection/
└── src/
    └── weight/
        ├── all_weight/ ← (PLACE HERE)
        │   └── trainX
        │       ├── best.pt
        │       └── last.pt
        │   
        └── yolov8m.pt

⚠️ Before running the project, make sure you have created the data.yaml file in the correct path as specified in section 4️⃣ Prepare dataset and config, and paste the following content:

• 📁 Path: .\Traffic-Sign-Detection\data\dataset\data.yaml

Traffic-Sign-Detection/
└── data/
    └── dataset/
        └── data.yaml ← (PLACE HERE)

• 📄 data.yaml

train: ./train/images
val: ./val/images
test: ./test/images

nc: 12
names: ['i.423.b', 'p.102', 'p.106.b', 'p.130', 'p.131.a', 'r.308.b', 'sus', 'w.201.a', 'w.203.c', 'w.207.b', 'w.207.c', 'w.209']

descriptions: [
  'Đường kẻ dành cho người đi bộ',                        # i.423.b
  'Cấm đi ngược chiều',                                   # p.102
  'Cấm xe tải trên N tấn',                                # p.106.b
  'Cấm dừng và đỗ xe',                                    # p.130
  'Cấm đỗ xe',                                            # p.131.a
  'Tuyến đường cầu vượt cắt qua',                         # r.308.b
  'TEXT SIGN',                                            # sus
  'Cảnh báo khúc cua nguy hiểm bên trái',                 # w.201.a
  'Cảnh báo đường hẹp',                                   # w.203.c
  'Cảnh báo giao nhau với đường không ưu tiên bên phải',  # w.207.b
  'Cảnh báo giao nhau với đường không ưu tiên bên trái',  # w.207.c
  'Cảnh báo đến khu vực đèn tín hiệu giao thông'          # w.209
]

roboflow:
  workspace: traffic-sign-detection-hzj9m
  project: traffic-sign-detection-e1j1v
  version: 4
  license: CC BY 4.0
  url: https://universe.roboflow.com/traffic-sign-detection-hzj9m/traffic-sign-detection-e1j1v/dataset/4

6️⃣ Training

• Ensure you have a valid dataset YAML. The project auto-detects in this order: data/dataset/data.yaml or .yml, then data/data.yaml or .yml, then repo root. You can override with env var DATA_YAML.
• Open Terminal and run this script from the repo root using module mode:

python -m src.training.train

7️⃣ Detect traffic signs

Option 1: Detect with images

• Place the traffic signs images you want to detect into the 📁input folder: .\Traffic-Sign-Detection\data\input
• Open Terminal and run the detection script from the repo root:

python -m src.pipeline.predict

Option 2: Detect with real time webcam/camera

• Open Terminal and run the script for basic real-time detection:

python -m src.pipeline.real_time_predict

Option 3: Detect with real time webcam with advanced smooth

• Make sure you have placed sort.py in the src directory.
• Open Terminal and run the script for advanced real-time detection with SORT:

python -m src.pipeline.real_time_predict_smooth_advanced

Option 4: Real-Time Detection with Vintern NLP Hybrid Model

• This option enables real-time traffic sign detection combined with NLP-based semantic analysis using the Vintern model hosted on Hugging Face.

• ⚙️ Setup Instructions

1. Create a .env file in the project’s root directory.

2. Go to Hugging Face Tokens and generate a new Access Token.

3. Add your token to the .env file as follows:

   HF_TOKEN=your_huggingface_token_here

• Open Terminal and run the script for advanced real-time detection with NLP Model:

python -m src.pipeline.real_time_predict_nlp_hybrid

8️⃣ Data

• Detected images: .\Traffic-Sign-Detection\data\output
• Detected videos: .\Traffic-Sign-Detection\data\real_time_output
• Trained weights: .\Traffic-Sign-Detection\weight\all_weight

9️⃣ Enjoy🎉

• Thank you for checking out our project! Feel free to explore, improve, or contribute 🚀

---

🌐 Connect with us

---

📌 Documents

	Project Report (PDF)
	Project Report (Word)
	Project Presentation (PPTX)