This paper explores a novel weakly supervised learning (WSL) framework for semantic segmentation of mobile laser scanning (MLS) point clouds. Specifically, a category balanced random annotation (CBRA) strategy is employed to obtain balanced labels and enhance model performance. Next, a deep consistency-guided self-distillation (DCS) mechanism is developed by transferring knowledge among different deep networks to exploit semantically rich information. By using only 0.1% labelled points, the proposed WSL framework achieved a competitive performance on three different point cloud datasets.

This study assesses the impact of oblique imagery and flight-planning scenarios (top-right image) on reducing height errors in UAV photogrammetric blocks. It explores the efficiency of combining nadir and oblique images, georeferenced by direct and indirect methods. Results shown in the bottom image demonstrate that integrating oblique images at the block centre with nadir images improves block accuracy, achieving 50% and 40% accuracy enhancements for direct and indirect georeferenced blocks, respectively.