ProtoGuard-Guided PROPEL: Class-Aware Prototype Enhancement and Progressive Labeling for Incremental 3D Point Cloud Segmentation

Abstract

3D point cloud semantic segmentation technology has been widely used in robotic navigation. Considering that the environment is evolving in real-world applications, offline-trained segmentation models may face the problem of catastrophic forgetting of previously seen classes. This work tailors class-incremental learning (CIL) method to address the problem of catastrophic forgetting in 3D point cloud segmentation. Regarding point clouds, we observe high similarity and unclear boundaries between different classes, and that class distribution is imbalanced. These lead to issues including misclassification between similar classes and the long-tailed problem, which have not been adequately addressed in previous CIL methods. Motivated by these, we propose the approach of ProtoGuard-guided PROPEL (Progressive Refinement Of PsEudo-Labels). Specifically, in the base-class training phase, ProtoGuard maintains geometric and semantic prototypes for each class, which are combined into prototype features using an attention mechanism. In the novel-class training phase, PROPEL inherits the base feature extractor and classifier, guiding pseudo-label propagation and updates based on density distribution and semantic similarity. Extensive experiments show that our approach achieves remarkable results on both the S3DIS and ScanNet datasets, improving the mIoU of 3D point cloud segmentation by a maximum of 20.39% under the 5-step CIL scenario on S3DIS.