Our Dataset
Public Dataset
We started with publicly available road datasets containing labeled images of potholes and speed bumps from various sources.
- 10,000+ images from open repositories
- Various lighting and weather conditions
- Multiple road types (urban, rural, highways)
Our Custom Dataset
Each team member contributed by collecting and labeling images from their local areas to enhance diversity.
- 5,000+ new images collected locally
- Each town's unique road conditions captured
- Seasonal variations included
Annotation Process
Roboflow
Used for consistent labeling across all images
Bounding Boxes
Precise annotation of potholes and speed bumps
Augmentation
Applied transformations to increase diversity
Final Merged Dataset
After careful processing and validation, we combined both datasets to create a robust training set.
Dataset Statistics
- 15,000+ total images
- 25,000+ labeled potholes
- 8,000+ labeled speed bumps
- Class balance maintained
Split
- 70% training set
- 20% validation set
- 10% test set
