Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval
Existing class-incremental learning methods have achieved significant progress in unimodal domains, but they face severe challenges of catastrophic forgetting and cross-modal interference when handling multimodal data. We propose a causal intervention-based multimodal class-incremental learning fram...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11014104/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849695397503565824 |
|---|---|
| author | Qiang Li Qiu-Yang Ma Ning Zhang Wei-Zhi Nie |
| author_facet | Qiang Li Qiu-Yang Ma Ning Zhang Wei-Zhi Nie |
| author_sort | Qiang Li |
| collection | DOAJ |
| description | Existing class-incremental learning methods have achieved significant progress in unimodal domains, but they face severe challenges of catastrophic forgetting and cross-modal interference when handling multimodal data. We propose a causal intervention-based multimodal class-incremental learning framework to address the stability-plasticity trade-off in 3D model classification and retrieval tasks. By employing backdoor adjustment-based causal intervention, our approach decouples historical task biases from new knowledge acquisition. It introduces an alternating modality freezing strategy to isolate parameter updates between geometric (point cloud) and visual (multi-view) modalities, thereby suppressing cross-modal gradient interference. Furthermore, a dynamic feature anchoring loss is utilized to constrain feature space drift and maintain cross-task consistency. Experimental results on the ModelNet40 and ShapeNetCore55 datasets demonstrate that our method achieves state-of-the-art performance compared to competitive baselines, confirming the effectiveness of multimodal redundancy in mitigating unimodal degradation. This work provides theoretical support and technical implementation pathways for lifelong learning systems in complex 3D scenarios. |
| format | Article |
| id | doaj-art-52f8cc3a0a0641e8b6d97d9c5215ecf7 |
| institution | DOAJ |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-52f8cc3a0a0641e8b6d97d9c5215ecf72025-08-20T03:19:47ZengIEEEIEEE Access2169-35362025-01-0113918469185910.1109/ACCESS.2025.357304111014104Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and RetrievalQiang Li0https://orcid.org/0000-0001-7129-1456Qiu-Yang Ma1https://orcid.org/0009-0001-8929-963XNing Zhang2Wei-Zhi Nie3https://orcid.org/0000-0002-0578-8138School of Microelectronics, Tianjin University, Tianjin, ChinaSchool of Microelectronics, Tianjin University, Tianjin, ChinaSchool of Information Resource Management, Renmin University of China, Beijing, ChinaSchool of Electrical and Information Engineering, Tianjin University, Tianjin, ChinaExisting class-incremental learning methods have achieved significant progress in unimodal domains, but they face severe challenges of catastrophic forgetting and cross-modal interference when handling multimodal data. We propose a causal intervention-based multimodal class-incremental learning framework to address the stability-plasticity trade-off in 3D model classification and retrieval tasks. By employing backdoor adjustment-based causal intervention, our approach decouples historical task biases from new knowledge acquisition. It introduces an alternating modality freezing strategy to isolate parameter updates between geometric (point cloud) and visual (multi-view) modalities, thereby suppressing cross-modal gradient interference. Furthermore, a dynamic feature anchoring loss is utilized to constrain feature space drift and maintain cross-task consistency. Experimental results on the ModelNet40 and ShapeNetCore55 datasets demonstrate that our method achieves state-of-the-art performance compared to competitive baselines, confirming the effectiveness of multimodal redundancy in mitigating unimodal degradation. This work provides theoretical support and technical implementation pathways for lifelong learning systems in complex 3D scenarios.https://ieeexplore.ieee.org/document/11014104/Class-incremental learningmultimodalcausal interventionknowledge distillation |
| spellingShingle | Qiang Li Qiu-Yang Ma Ning Zhang Wei-Zhi Nie Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval IEEE Access Class-incremental learning multimodal causal intervention knowledge distillation |
| title | Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval |
| title_full | Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval |
| title_fullStr | Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval |
| title_full_unstemmed | Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval |
| title_short | Causal Intervention-Based Multimodal Class-Incremental Learning Network for 3D Model Classification and Retrieval |
| title_sort | causal intervention based multimodal class incremental learning network for 3d model classification and retrieval |
| topic | Class-incremental learning multimodal causal intervention knowledge distillation |
| url | https://ieeexplore.ieee.org/document/11014104/ |
| work_keys_str_mv | AT qiangli causalinterventionbasedmultimodalclassincrementallearningnetworkfor3dmodelclassificationandretrieval AT qiuyangma causalinterventionbasedmultimodalclassincrementallearningnetworkfor3dmodelclassificationandretrieval AT ningzhang causalinterventionbasedmultimodalclassincrementallearningnetworkfor3dmodelclassificationandretrieval AT weizhinie causalinterventionbasedmultimodalclassincrementallearningnetworkfor3dmodelclassificationandretrieval |