Diffuse&Refine: Intrinsic Knowledge Generation and Aggregation for Incremental Object Detection

Wang, Jianzhou; Wu, Yirui; Yuan, Lixin; Zhang, Wenxiao; Liu, Jun; Chen, Junyang; Wang, Huan; Wang, Wenhai

doi:10.24963/ijcai.2025/700

Diffuse&Refine: Intrinsic Knowledge Generation and Aggregation for Incremental Object Detection

Jianzhou Wang, Yirui Wu, Lixin Yuan, Wenxiao Zhang, Jun Liu, Junyang Chen, Huan Wang, Wenhai Wang

Download BibTeX Cite

Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence. Main Track. Pages 6289-6297.

https://doi.org/10.24963/ijcai.2025/700

Keywords

Machine Learning: ML: Incremental learning
Computer Vision: CV: Recognition (object detection, categorization)

Abstract

Incremental Object Detection(IOD) targets at progressively extending capability of object detectors to recognize new classes. However, representation confusion between old and new classes leads to catastrophic forgetting. To alleviate this problem, we propose DiffKA, with intrinsic knowledge generated and aggregated by forward and backward diffusion, gradually establishing rigid class boundary. With incremental streaming data, forward diffusion spreads information to generate potential inter-class associations among new- and old-class prototypes within a hierarchical tree, named as Intrinsic Correlation Tree(ICTree), to store intrinsic knowledge. Afterwards, backward diffusion refines and aggregates the generated knowledge in ICTree, explicitly establishing rigid class boundary to mitigate representation confusion. To keep semantic consistency with extreme IOD settings, we reorganize semantic relevance of old- and new-class prototypes in paradigms to adaptively and effectively update DiffKA. Experiments on MS COCO dataset show DiffKA achieves state-of-the-art performance on IOD tasks with significant advantages.

Jianzhou Wang, Yirui Wu, Lixin Yuan, Wenxiao Zhang, Jun Liu, Junyang Chen, Huan Wang, and Wenhai Wang. 2025. Diffuse&Refine: Intrinsic Knowledge Generation and Aggregation for Incremental Object Detection. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25. IJCAI, 6289–6297. https://doi.org/10.24963/ijcai.2025/700

@inproceedings{ijcai2025p700,
  title     = {Diffuse&Refine: Intrinsic Knowledge Generation and Aggregation for Incremental Object Detection},
  author    = {Wang, Jianzhou and Wu, Yirui and Yuan, Lixin and Zhang, Wenxiao and Liu, Jun and Chen, Junyang and Wang, Huan and Wang, Wenhai},
  booktitle = {Proceedings of the Thirty-Fourth International Joint Conference on
               Artificial Intelligence, {IJCAI-25}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {James Kwok},
  pages     = {6289--6297},
  year      = {2025},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2025/700},
  url       = {https://doi.org/10.24963/ijcai.2025/700},
}

TY  - CPAPER
TI  - Diffuse&Refine: Intrinsic Knowledge Generation and Aggregation for Incremental Object Detection
AU  - Wang, Jianzhou
AU  - Wu, Yirui
AU  - Yuan, Lixin
AU  - Zhang, Wenxiao
AU  - Liu, Jun
AU  - Chen, Junyang
AU  - Wang, Huan
AU  - Wang, Wenhai
T2  - Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, IJCAI-25
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2025
SP  - 6289
EP  - 6297
DO  - 10.24963/ijcai.2025/700
UR  - https://doi.org/10.24963/ijcai.2025/700
ER  -