Multimodal material segmentation (MCubeS) dataset contains 500 sets of images from 42 street scenes. The dataset provides annotated ground truth labels for both material and semantic segmentation for every pixel.
10 PAPERS • 1 BENCHMARK