[CVPR2026] Vision-Language Attribute Disentanglement and Reinforcement
for Lifelong Person Re-Identification

Kunlun Xu¹  Haotong Cheng¹  Jiangmeng Li²  Xu Zou³  Jiahuan Zhou^1*

¹Wangxuan Institute of Computer Technology, Peking University  |  ²University of Chinese Academy of Sciences  |  ³School of Artificial Intelligence and Automation, Huazhong University of Science and Technology

---

Official repository for Vision-Language Attribute Disentanglement and Reinforcement for Lifelong Person Re-Identification (CVPR 2026).

Overview

VLADR tackles Lifelong Person Re-Identification using vision-language alignment. The pipeline consists of two stages:

• Stage 1 — Multi-grain Text Attribute Disentanglement (MTAD): Extracts multi-grain attributes via local attribute extraction and global attribute modeling.
• Stage 2 — Inter-domain Cross-modal Attribute Reinforcement (ICAR): Loads Stage 1 prompt checkpoints and leverages pre-extracted text descriptions to fine-tune the image encoder.

Installation

After project VLADR have been downloaded to home, then run:

conda create -n VLADR python=3.9
conda activate VLADR
pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124
pip install -r requirements.txt
python setup.py develop

After setup, activate the environment normally in future sessions with conda activate VLADR.

Prepare Datasets

Download the following person re-identification datasets and place them under a root directory (e.g., PRID/):

Dataset	Link
Market-1501	Google Drive
MSMT17	pkuvmc.com
CUHK03	GitHub
SenseReID	Google Drive
Others	Torchreid Docs / light-reid

The expected directory structure:

PRID
├── CUHK01/
├── CUHK02/
├── CUHK03/
├── CUHK-SYSU/
├── DukeMTMC-reID/
├── grid/
├── i-LIDS_Pedestrain/
├── MSMT17_V2/
├── Market-1501/
├── prid2011/
├── SenseReID/
└── viper/

Quick Start

Training

Stage 1 — Multi-grain Text Attribute Disentanglement (MTAD)

bash train1.sh

Attributes have been pre-extracted and stored in ./_BLIP_TEXT_DESC and _STAGE1_PROMPTS_WEIGHT. You can directly proceed to Stage 2 simply by using the aforementioned resources.

Stage 2 — Inter-domain Cross-modal Attribute Reinforcement (ICAR)

bash train2.sh

Evaluation

Pre-trained models are provided for quick evaluation:

bash test.sh

Results

Results obtained with two NVIDIA RTX 4090 GPUs:

Citation

If you find this work useful for your research, please consider citing:

@inproceedings{xu2026vladr,
  title={Vision-Language Attribute Disentanglement and Reinforcement for Lifelong Person Re-Identification},
  author={Xu, Kunlun and Cheng, Haotong and Li, Jiangmeng and Zou, Xu and Zhou, Jiahuan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  year={2026}
}

We have conducted a series of research in Lifelong Person Re-Identification as follows.

Semi-Supervised Lifelong Learning

@inproceedings{xu2025self,
  title={Self-reinforcing prototype evolution with dual-knowledge cooperation for semi-supervised lifelong person re-identification},
  author={Xu, Kunlun and Zhuo, Fan and Li, Jiangmeng and Zou, Xu and Zhou, Jiahuan},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={3564--3574},
  year={2025}
}

Image-level Distribution Modeling and Transfer:

@inproceedings{xu2025dask,
  title={Dask: Distribution rehearsing via adaptive style kernel learning for exemplar-free lifelong person re-identification},
  author={Xu, Kunlun and Jiang, Chenghao and Xiong, Peixi and Peng, Yuxin and Zhou, Jiahuan},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={39},
  number={9},
  pages={8915--8923},
  year={2025}
}

Feature-level Distribution Modeling and Prototyping:

@article{zhou2025distribution,
  title={Distribution-Aware Knowledge Aligning and Prototyping for Non-Exemplar Lifelong Person Re-Identification},
  author={Zhou, Jiahuan and Xu, Kunlun and Zhuo, Fan and Zou, Xu and Peng, Yuxin},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2025},
  publisher={IEEE}
}

@inproceedings{xu2024distribution,
  title={Distribution-aware Knowledge Prototyping for Non-exemplar Lifelong Person Re-identification},
  author={Xu, Kunlun and Zou, Xu and Peng, Yuxin and Zhou, Jiahuan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={16604--16613},
  year={2024}
}

Long Short-Term Knowledge Rectification and Consolidation:

@article{xu2025long,
  title={Long Short-Term Knowledge Decomposition and Consolidation for Lifelong Person Re-Identification},
  author={Xu, Kunlun and Liu, Zichen and Zou, Xu and Peng, Yuxin and Zhou, Jiahuan},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2025},
  publisher={IEEE}
}


@inproceedings{xu2024lstkc,
  title={Lstkc: Long short-term knowledge consolidation for lifelong person re-identification},
  author={Xu, Kunlun and Zou, Xu and Zhou, Jiahuan},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={38},
  number={14},
  pages={16202--16210},
  year={2024}
}

Lifelong Learning with Label Noise:

@inproceedings{xu2024mitigate,
  title={Mitigate Catastrophic Remembering via Continual Knowledge Purification for Noisy Lifelong Person Re-Identification},
  author={Xu, Kunlun and Zhang, Haozhuo and Li, Yu and Peng, Yuxin and Zhou, Jiahuan},
  booktitle={Proceedings of the 32nd ACM International Conference on Multimedia},
  pages={5790--5799},
  year={2024}
}

Prompt-guided Adaptive Knowledge Consolidation:

@article{li2024exemplar,
  title={Exemplar-Free Lifelong Person Re-identification via Prompt-Guided Adaptive Knowledge Consolidation},
  author={Li, Qiwei and Xu, Kunlun and Peng, Yuxin and Zhou, Jiahuan},
  journal={International Journal of Computer Vision},
  pages={1--16},
  year={2024},
  publisher={Springer}
}

Compatible Lifelong Learning:

@inproceedings{cui2024learning,
  title={Learning Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification},
  author={Cui, Zhenyu and Zhou, Jiahuan and Wang, Xun and Zhu, Manyu and Peng, Yuxin},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={16614--16623},
  year={2024}
}

Acknowledgement

Our code is built upon DASK and CLIP-ReID. We sincerely thank the authors for their excellent work.

Contact

For questions, feel free to reach out at xkl@stu.pku.edu.cn.

Visit our lab homepage OV³ Lab for more papers, code, and datasets.