Munzur University-driving Labeled Video Dataset (MUVid)

This dataset was collected within the scope of the TÜBİTAK 1002-A project numbered 124E787 from the campus areas of Munzur University and is named The Munzur University-driving Labeled Video Dataset (MUVid). The MUVid dataset was recorded in real driving conditions in the campus of Munzur University, located in Tunceli, Türkiye, using a Mevo Start camera with a resolution of 1280 × 720, which was internally mounted on the windshield of a vehicle. The dataset was created for semantic segmentation of drivable road areas and lane line estimation tasks in autonomous driving applications.

To capture different illumination conditions, the recordings were specifically conducted during twilight hours close to sunset. No adverse weather conditions or environmental variations were present during data collection. The recorded video data was temporally sampled into frames by extracting one image every 3 seconds, resulting in a total of 307 RGB images. These images were manually annotated using the CVAT annotation tool into two pixel-wise classes: background and drivable road area.

The images in the dataset are numbered from 0 to 306, and each corresponding mask follows the same naming convention. The images and masks are stored in JPG and PNG formats, respectively. The dataset is organized in a structured manner, where images are placed in the “img” folder and their corresponding masks are stored in the “mask” folder. This organization ensures that the dataset is cleanly structured and directly suitable for machine learning and deep learning applications.