Welcome to the official GitHub organization of PKU OV³ Lab at the Wangxuan Institute of Computer Technology, Peking University.

We work on open-environment visual understanding and multimodal intelligence, with a focus on building vision systems that can adapt, generalize, and keep learning under realistic data shifts. Our repositories host official implementations of our papers, research prototypes, and reproducible resources for the community.

Our current interests include:

• Open-Environment Computer Vision: lifelong person re-identification, open-world recognition, robust visual perception, and vision systems under distribution shifts.
• Efficient Learning from Limited Data: zero-shot and few-shot learning, incremental learning, online learning, noisy-label learning, prompt learning, and test-time adaptation.
• Multi-Modal Learning: vision-language learning, vision-language-action models, cross-modal transfer, visible-infrared learning, sketch-to-photo understanding, and 2D-to-3D perception.
• AIGC and AI4Science: generative models, foundation-model adaptation, and AI-driven scientific discovery.

• CVPR2026-PDF: Test-Time Perturbation Tuning with Delayed Feedback for Vision-Language-Action Models.
• CVPR2026-VLADR: Vision-Language Attribute Disentanglement and Reinforcement for Lifelong Person Re-Identification.
• ICLR2026-N2L: Naming to Learn for class-incremental learning with vision-language models and unlabeled data.

• NeurIPS2025-SSP: State Space Prompting for video understanding.
• ICML2025-GAPrompt: Geometry-Aware Point Cloud Prompt for 3D vision models.
• ICML2025-VGP: Vision Graph Prompting via semantic low-rank decomposition.
• ICCV2025-UPP: Unified Point-Level Prompting for robust point cloud analysis.
• AAAI2025-CAPrompt: Cyclic Prompt Aggregation for pre-trained model based class-incremental learning.

• NeurIPS2025-C2Prompt: Class-aware client knowledge interaction for federated continual learning.
• NeurIPS2025-KFF: Class-aware domain knowledge fusion and fission for continual test-time adaptation.
• CVPR2024-FCS: Feature Calibration and Separation for non-exemplar class-incremental learning.
• MM2024-PPE: Progressive Prototype Evolving for dual-forgetting mitigation in non-exemplar online continual learning.

• CVPR2024-DKP: Distribution-aware Knowledge Prototyping for non-exemplar lifelong person re-identification.
• TPAMI-DKP_Plus_Plus: Distribution-aware Knowledge Aligning and Prototyping for lifelong person re-identification.
• AAAI2024-LSTKC: Long Short-Term Knowledge Consolidation for lifelong person re-identification.
• IJCV2024-PAEMA: Exemplar-free lifelong person re-identification via prompt-guided adaptive knowledge consolidation.
• CVPR2025-DKC: Differentiated Knowledge Consolidation for cloth-hybrid lifelong person re-identification.

• CVPR2026-RS-SSM: Refining Forgotten Specifics in State Space Model for video semantic segmentation.
• CVPR2025-STOP: Integrated Spatial-Temporal Dynamic Prompting for video understanding.
• CVPR2025-SCAP: Transductive test-time adaptation via supportive clique-based attribute prompting.

Most repositories in this organization are named by venue and year, such as CVPR2026-*, ICCV2025-*, ICML2025-*, and AAAI2025-*. Please visit each repository for installation instructions, pretrained models, datasets, and citation information.

If you use our code or find our work helpful, please cite the corresponding paper listed in the repository README.

We are actively looking for motivated postdocs, Ph.D. students, master students, and research interns interested in computer vision, multimodal learning, AI4Science, and related areas.

For more information, please visit:

• Lab GitHub: github.com/PKU-OV3-LAB
• Prof. Jiahuan Zhou's homepage: zhoujiahuan1991.github.io

For academic collaboration, internship opportunities, and questions about specific repositories, please contact Prof. Jiahuan Zhou or open an issue in the corresponding project repository.