RGB Road Scene Material Segmentation

Material informed robotics - spectral perception for object identification and parameter inference

“…[93] explores the use of RGB imagery alone to perform material segmentation with transformbased neural networks. They also propose a dataset called KITTI-Materials, containing 1000 frames with 20 different material categories [93]. The same group also augmented RGB imagery with NIR and polarization images to improve classification accuracy on materials such as metal and water [94].…”