Pothole Detection with YOLOv8

View Project: pothole-detect.streamlit.app

This repository contains the complete workflow for training a YOLOv8s object detection model to identify potholes in road images. The project covers the full pipeline: sourcing raw images, performing manual data labeling, and training the model.

Project Overview

The goal of this project was to learn the end-to-end process of building a custom object detector. The model is trained on a single class: pothole.

Workflow and Tools

This project was built using the following components:

• Model: YOLOv8s (small variant) by Ultralytics.
• Data Sourcing: Raw images were taken from the Pothole Image Segmentation Dataset on Kaggle.
• Data Labeling: Manual bounding box annotation was performed using Label Studio. The labels were then exported in YOLO .txt format.
• Training: The model was trained using a free GPU on Google Colab. The notebook used is based on the Train YOLO Models template.

How to Use

The main training process is documented in the included Jupyter Notebook. To run the training, you will need to:

1. Have your own labeled dataset (images and YOLO-formatted labels) uploaded to a service like Google Drive.
2. Create a data.yaml file that points to your training and validation data paths.
3. Run the cells in the notebook to install dependencies, mount your drive, and begin the training.