In the 3D point cloud semantic segmentation task , using a small amount of labeled point cloud data for semantic segmentation can save the cost of human labeling , and has attracted widespread attention from the academic community in recent years. Traditional 3D point cloud semantic segmentation methods mostly use fully supervised methods , which often require manpower and time to label a large number of point clouds , while using weakly super- vised methods only requires a small amount of labeling on point clouds to achieve the same purpose as fully supervised methods. This paper reviews and discusses the development of weakly supervised semantic segmentation of 3D point cloud in recent years , and summarizes the related methods of weakly supervised semantic segmentation from different perspectives. Based on these methods , the results are quantitatively analyzed and discussed on four public datasets. Finally , the challenges of weakly supervised semantic segmentation of 3D point cloud are summarized , and the future development direction is prospected.