Weakly supervised semantic segmentation for point cloud based on view-based adversarial training and self-attention fusion

Yongwei Miao, Guoxiang Ren, Jin-Rong Wang, Fuchang Liu

Abstract: Highlights•We propose effective data augmentation methods, including view resampling and adversarial training, to tackle partial annotations.•To fully integrate the global and local features of point cloud, we employ high-dimensional feature self-attention mechanism and multi-layer perceptrons to fuse multi-level features.•We present a novel point cloud segmentation framework, which can achieve the higher segmentation accuracy than traditional segmentation networks for different point cloud scenes.•Extensive experiments demonstrate the robustness, effectiveness, and generalization of the proposed weakly supervised point cloud semantic segmentation network.