Pothole Image Segmentation Dataset
Annotated image dataset for the detection and segmentation of potholes on roads, with pre-processing and augmentation to improve the robustness of the models.
780 annotated images in Yolov8 format, 640x640 px, with sets train (720) and validation (60)
Apache 2.0
Description
The “Pothole Image Segmentation Dataset” dataset contains 780 images annotated in Yolov8 format, focused on the detection and segmentation of potholes for road safety. The images are resized to 640x640 pixels and increased to enrich the training data.
What is this dataset for?
- Train object segmentation models to detect potholes
- Develop automatic road monitoring and maintenance systems
- Improving road safety through the rapid and accurate detection of damage
Can it be enriched or improved?
This dataset can be enriched by adding images of different road conditions, other types of degradation, or by integrating additional annotations (e.g. finer segmentation, multiple categories). Increasing data and adding GPS metadata could also be considered.
🔎 In summary
🧠 Recommended for
- Computer vision researchers
- Smart City Solution Developers
- Road safety engineers
🔧 Compatible tools
- Roboflow
- YoloV8
- Detectron2
- TensorFlow
- PyTorch
💡 Tip
Exploit built-in increases to improve the robustness of the model in the face of various lighting conditions and perspectives.
Frequently Asked Questions
What annotation format is used in this dataset?
The dataset uses the YoloV8 format for pothole segmentation annotations.
How many images are dedicated to training and validation?
720 images are for training and 60 for validation.
Can this dataset be used for other types of road object detection?
Yes, with additional annotations, it could be adapted to detect other faults or objects on the road.
