Cosine Prompt-Based Class Incremental Semantic Segmentation for Point Clouds
Open Access
- 16 October 2025
- journal article
- research article
- Published by MDPI AG in Algorithms
- Vol. 18 (10) , 648
Abstract
Although current 3D semantic segmentation methods have achieved significant success, they suffer from catastrophic forgetting when confronted with dynamic, open environments. To address this issue, class incremental learning is introduced to update models while maintaining a balance between plasticity and stability. In this work, we propose CosPrompt, a rehearsal-free approach for class incremental semantic segmentation. Specifically, we freeze the prompts for existing classes and incrementally expand and fine-tune the prompts for new classes, thereby generating discriminative and customized features. We employ clamping operations to regulate backward propagation, ensuring smooth training. Furthermore, we utilize the learning without forgetting loss and pseudo-label generation to further mitigate catastrophic forgetting. We conduct comparative and ablation experiments on the S3DIS dataset and ScanNet v2 dataset, demonstrating the effectiveness and feasibility of our method.Keywords
Funding Information
- Natural Science Foundation of Shanxi Province (202203021212138)
- National Natural Science Foundation of China (62272426)
- Foundation of Shanxi Key Laboratory of Machine Vision and Virtual Reality (447-110103)
This publication has 23 references indexed in Scilit:
- Balanced residual distillation learning for 3D point cloud class-incremental semantic segmentationExpert Systems with Applications, 2025
- Class-Incremental Learning: A SurveyIEEE Transactions on Pattern Analysis and Machine Intelligence, 2024
- A comprehensive overview of deep learning techniques for 3D point cloud classification and semantic segmentationMachine Vision and Applications, 2024
- A Comprehensive Survey of Continual Learning: Theory, Method and ApplicationIEEE Transactions on Pattern Analysis and Machine Intelligence, 2024
- Continual learning on 3D point clouds with random compressed rehearsalComputer Vision and Image Understanding, 2023
- Continual Semantic Segmentation via Structure Preserving and Projected Feature Alignment2022
- L3DOC: Lifelong 3D Object ClassificationIEEE Transactions on Image Processing, 2021
- PODNet: Pooled Outputs Distillation for Small-Tasks Incremental Learning2020
- Learning without ForgettingIEEE Transactions on Pattern Analysis and Machine Intelligence, 2017
- Overcoming catastrophic forgetting in neural networksProceedings of the National Academy of Sciences of the United States of America, 2017