pothole-detection

Training a computer vision task to detect and segment potholes in images. This model was trained on the pothole-600 dataset.

Datasets Used

• Nienaber 1
• Nienaber 2
• Road Damage Detection
• Brazilian NDTI

Setup

Create and Activate Virtual Environment

python3 -m venv venv --system-site-packages
source venv/bin/activate

Install required packages

pip install -r requirements.txt

Model Testing

Model trained on an augmented dataset containing 6600 images using a Feature Pyramid Network. This model was able to achive an IOU of 0.86