OPOCA: One Point One Class Annotation for LiDAR Point Cloud Semantic Segmentation

Weijie Huang, Pufan Zou, Yan Xia, Chenglu Wen, Yu Zang, Cheng Wang, Guoqing Zhou

Abstract: This article tackles the problem of requiring a large amount of data annotation in the LiDAR point cloud semantic segmentation (PCSS) task by proposing OPOCA, a weakly supervised network that only annotates one point per class in a single LiDAR scan. To compensate for the supervisory losses due to extremely few annotated labels, a large number of pseudo labels is first generated using a pseudo label spreading module (PLSM), whereas the potential ambiguity and inaccuracy are further addressed by a carefully-designed spread distance loss (SDL) and a range image auxiliary module (RIAM). Moreover, we propose an iterative self-training module (STM) to increase the high-quality pseudo labels for the next round of training. Extensive experiments on various benchmark datasets (SemanticKITTI, Waymo Open Dataset (WOD), and SemanticPOSS) demonstrate the rationality of each module and the superior performance of the proposed network over the current baseline with below 0.1‰ labels.