Overview:
This pothole detection project utilizes YOLOv8m to identify and classify potholes in road conditions. The model processes a dataset of 665 images to accurately detect potholes and help improve road maintenance and safety. After 50 training epochs with rotation augmentation, the model has demonstrated reliable detection performance.
Key Features:
1. YOLOv8m for Pothole Detection:
The model is specifically designed to detect potholes in road conditions, providing fast and reliable object detection performance.
2. Dataset of 665 Images:
The model was trained on a dataset of 665 images, each with a resolution of 640×640, focused solely on the “Pothole” class to optimize detection accuracy.
3. Augmentation with Rotation:
Rotation-based augmentation was applied to enhance the model’s ability to detect potholes from different angles and perspectives, improving robustness.4.
4. Model Performance:
Achieved a mean average precision (mAP) of 0.721, precision of 0.714, and recall of 0.694 after 50 epochs, indicating reliable but improvable results.
5. Potential Improvements:
Increasing the number of epochs and adding more augmentations could further refine the model’s accuracy, as the loss trend suggests room for growth.
Category:
Tags:
Links
Portfolio 
Leave a comment