Skip to content

yujimin413/Road-Damage-Detection

 
 

Repository files navigation

🖤 CRACKPINK: Road Defect Detection with YOLOv12 🩷

🧠 YOLO Object Detection - Performance Analysis Report

This repository presents the performance evaluation of a YOLOv12 object detection model trained on custom-labeled road surface data for identifying various crack types.


📁 Experimental Setup

📌 Model Configuration

  • Model: YOLOv12 (Extra-Large variant)
  • Input Size: 640 × 640 pixels
  • Epochs: 150
  • Optimizer: SGD (Stochastic Gradient Descent)
  • Learning Rate: 0.001
  • Confidence Threshold: 0.10
  • IoU Threshold: 0.15

🧪 Data Augmentation

To enhance robustness and generalization, the following augmentation techniques were applied:

  • Random scaling
  • Shearing transformations
  • Gaussian noise addition
  • Random cropping
  • Rotation (±10°)

📸 Sample Predictions

KakaoTalk_Photo_2025-07-26-18-16-32
KakaoTalk_Photo_2025-07-26-18-16-36

📊 Observations & Insights

🔍 Key Hyperparameter Findings

  • IoU Threshold (0.15) provided a balanced trade-off between precision and recall.
  • 150 training epochs allowed for sufficient convergence without overfitting.
  • Proper train/validation split significantly influenced model generalization.

🚀 YOLOv12 Advantages

Compared to previous versions, YOLOv12 showed improved performance due to:

  • Enhanced feature extraction capability
  • Refined neck architecture enabling better multi-scale feature fusion
  • Anchor-free detection head: reduced computational load with high accuracy

🎯 Augmentation Impact

The augmentation strategy notably boosted model robustness:

  • Enabled the model to generalize across different lighting and geometric distortions
  • Improved detection stability in real-world road conditions

🗒️ Notes

  • This report summarizes results without any ensemble techniques.
  • Future directions include:
    • Applying test-time augmentation
    • Exploring model ensembling to further enhance performance

About

Road damage detection model and application for 2025 IMSC Hackathon by USC

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 100.0%