3D Semantic Segmentation in the Wild: Learning Generalized Models for Adverse-Condition Point Clouds

Figure 1. We introduce SemanticSTF, an adverse-weather LiDAR point cloud dataset with dense point-level annotations that can be exploited for the study of point cloud semantic segmentation under all-weather conditions (including fog, snow, and rain). The graph on the left shows one scan sample captured on a snowy day, and the one on the right shows the corresponding point-level annotations.