<p>To address the issues of poor output consistency, information loss, and blurred boundaries caused by incomplete truth labeling in current indoor large-scale weakly-supervised point cloud semantic segmentation methods.&#xa0;We propose a weakly supervised semantic segmentation method for indoor large-scale point clouds based on consistency constraints and location guidance. Provide additional constraints on the input point cloud to learn to enhance the input consistency of the point cloud data, to better understand the essential characteristics of the data and improve the generalization ability of the model. In the process of point feature extraction, a structure-aware point cloud feature coding module is introduced to enhance the perception ability of the model, and a multi-scale attention path is constructed in it. By strengthening the channel dependence and enhancing the representation ability of important features, the model can predict the semantic labels of point clouds more effectively.&#xa0;The experimental results show that the proposed method can obtain better segmentation results when using 1.0 % truth labels for training on large-scale indoor datasets S3DIS and Scannet-v2. At the same time, it has achieved good results in the segmentation process of SemanticKITTI dataset and real point cloud.&#xa0;The method proposed in this paper has good segmentation performance for large-scale indoor scenes, and the method has good universality.</p>