Pothole Detection Dataset
This dataset contains images of roads with and without potholes, allowing the training of visual detection models for automated road maintenance. Although quite modest in volume, it is still useful for POCs or lightweight model tests.
681 JPG images divided into two classes: normal roads and roads with potholes
CC0: Public Domain
Description
The dataset Pothole Detection is composed of 681 images sorted into two categories: normal roads and roads with potholes. The images come from a variety of online sources, with a variety of viewpoints, and are particularly useful for road maintenance or autonomous driving projects.
What is this dataset for?
- Train a classification model or detection of road irregularities
- Create an embedded surveillance system for autonomous vehicles or drones
- Develop solutions to help urban maintenance
Can it be enriched or improved?
Yes, by adding precise annotations (bounding boxes), by integrating varied weather conditions or by capturing images with more geographic diversity. The addition of videos or time sequences would also be an asset for more complex models.
🔎 In summary
🧠 Recommended for
- Student projects
- Urban maintenance
- YOLO/MobileNet model testing
🔧 Compatible tools
- Roboflow
- TensorFlow
- PyTorch
- OpenCV
💡 Tip
Combine it with a GPS system or embedded camera for a real-time detection project on a smartphone or vehicle.
Frequently Asked Questions
Does this dataset contain pothole position annotations?
No, the images are organized by folder, but there are no bounding boxes or associated coordinates.
Is it enough to train a robust model?
The volume is modest but sufficient for prototypes or as a complement to other larger datasets.
Can it be used in a commercial or open-source context?
Yes, the CC0 license allows unrestricted commercial, open-source, or academic use.
