Pothole Detection and Severity Classification

This project implements a robust end-to-end system for detecting potholes and classifying their severity using deep learning. The final model integrates a detection stage (YOLOv11) with a dedicated severity classification model, designed to work together as a pipeline.

Final Model

The final model is a two-stage pipeline:

1. Detection using YOLOv11 (detection_best.pt)
2. Severity Classification using YOLOV8 (classifier_best.pt)

This modular architecture helps isolate detection from classification, allowing use of different training datasets and improving performance on severity estimation.

How to Run

Step-by-Step Instructions

Run Final Pipeline

• Open the pipeline notebook
• This executes the full model: detection followed by severity classification
• Outputs are saved and visualized

Additional Tools

Detection Model Training

• To train new YOLO models, use the detection Jupyter notebook
• This was used to test various YOLOv configurations
• This file needs to be run in Google Colab

Run Dataset Transformation

• Open and run the dataset_transformation notebook
• This loads and preprocesses data for the classification model training

Large Dataset Training

• The zipped folder (Pothole Detection.v1i.yolov11.zip) was used to train a larger pothole detection model on a more extensive dataset. Not uploaded in github due to large size. But can be loaded through roboflow. Look at detection for the code.

Results

• Realistic detection and severity classification on both clean and rainy conditions
• Visual examples available in detection_classification_results.png

Notes

• Severity labels: major, medium
• Designed to be modular, scalable, and realistic for deployment
• Trained and validated on data with real-world camera perspectives

Citations

@software{yolov8_ultralytics, author = {Glenn Jocher and Ayush Chaurasia and Jing Qiu}, title = {Ultralytics YOLOv8}, version = {8.0.0}, year = {2023}, url = {https://github.com/ultralytics/ultralytics}, orcid = {0000-0001-5950-6979, 0000-0002-7603-6750, 0000-0003-3783-7069}, license = {AGPL-3.0} }

@software{yolo11_ultralytics, author = {Glenn Jocher and Jing Qiu}, title = {Ultralytics YOLO11}, version = {11.0.0}, year = {2024}, url = {https://github.com/ultralytics/ultralytics}, orcid = {0000-0001-5950-6979, 0000-0002-7603-6750, 0000-0003-3783-7069}, license = {AGPL-3.0} }

@software{yolov5, title = {Ultralytics YOLOv5}, author = {Glenn Jocher}, year = {2020}, version = {7.0}, license = {AGPL-3.0}, url = {https://github.com/ultralytics/yolov5}, doi = {10.5281/zenodo.3908559}, orcid = {0000-0001-5950-6979} }

https://medium.com/data-science/the-comprehensive-guide-to-training-and-running-yolov8-models-on-custom-datasets-22946da259c3