Real-World Road Lane Segmentation Dataset with Depth-Derived LiDAR and Radar Proxy Representations

This dataset contains real-world monocular RGB road images collected from Andhra Pradesh and Tamil Nadu, India, along with manually annotated lane segmentation masks and depth-derived proxy representations analogous to LiDAR and radar inputs. The dataset includes approximately 4,300 RGB images, of which 1,081 are annotated with binary pixel-wise masks corresponding to lane markings. The annotated subset captures diverse road conditions, including variations in illumination, road surface quality, and lane visibility. In addition to RGB images and segmentation masks, the dataset provides depth maps generated using MiDaS monocular depth estimation. These depth maps are stored in NumPy (.npy) format and serve as simulated range cues to emulate LiDAR- and radar-like spatial information. The dataset does not contain measurements from physical LiDAR or radar sensors. The dataset is intended for research in lane detection, semantic segmentation, depth-aware feature fusion, and continual learning within vision-based autonomous driving systems.