Traffic Sign Detection

This project features a YOLOv8-based model for detecting and classifying traffic signs in images, integrated with a Gradio interface for interactive use. Upload an image to view detected traffic signs with bounding boxes and labels, powered by a model trained on the German Traffic Sign Recognition Benchmark (GTSRB) dataset. The model is deployed on Hugging Face Spaces at https://huggingface.co/spaces/salvirezwan/Traffic-Sign-Detection.

Usage

1. Upload an Image: Use the Gradio interface to upload an image containing traffic signs.
2. View Results: The app displays the image with bounding boxes and labels for detected traffic signs.
3. Confidence Threshold: Detections are filtered at a confidence score of 0.5.

Model Details

• Architecture: YOLOv8 nano
• Dataset: GTSRB, 43 classes, ~39,209 training images
• Training: 50 epochs, 640x640 image size, 80% training / 20% validation split
• Validation Metrics (final epoch): Details provided in results.csv file
• Platform: Trained in Google Colab with T4 GPU
• Model File: traffic_sign_detection.pt

Project Insights

• Challenges:
• Managed Google Colab runtime limits by saving checkpoints (e.g., epoch5.pt) to Google Drive for resuming training.
• Use Case: Suitable for real-time traffic sign detection in autonomous driving or driver assistance systems.

Deployment

• Hugging Face Space: Hosted at https://huggingface.co/spaces/salvirezwan/Traffic-Sign-Detection.
• Local Testing: Use python app.py with the required dependencies in requirements.txt.

Setup Instructions

1. Clone the repository:

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

2. Install dependencies:

pip install -r requirements.txt

3. Run the Gradio app:

python app.py

5. Upload an image via the local Gradio interface.

Future Improvements

• Retrain with augmented data (e.g., multiple signs, varied backgrounds) for better generalization.
• Explore YOLOv8 small or medium for improved accuracy.