Pothole Videos

Contains a video dataset of road potholes for semantic segmentation purposes. Data is taken from a point of view perpendicular to the road. Contains a total of 619 videos (image, mask) with an aspect ratio of 1:1, 1080x1080 resolution and 48 image frames for each video. The data was divided manually with a ratio of 60/20/20 with 372 for training, 124 for validation, and 123 for test. Ground truth is made manually using Adobe After Effects. The videos in this dataset were captured in Hulu Sungai Tengah Regency, Indonesia. If you use this dataset, please cite the following paper: M. Ihsan, M. A. Amrizal, and A. Harjoko, “A pothole video dataset for semantic segmentation,” Data in Brief, vol. 53, 2024.