HSI-Road Relabeled: Surface-Aware Road-Scene Segmentation
The HSI-Road dataset lacked surface labels, so this work introduced a six-class taxonomy (Background, Asphalt, Concrete, Dirt, Water, Grass) manually labeled using RGBori for better detail. A registration pipeline created a pseudo-NIR grayscale from three channels. Evaluation showed that combining registered RGB and NIR data significantly improves segmentation performance over using any single modality alone.