PAD

Official repository for “Prompt-Anchored Vision–Text Distillation for Lifelong Person Re-identification” (CVPR 2026).

Overview

We propose PAD (Prompt-Anchored Vision–Text Distillation), a framework for exemplar-free lifelong person re-identification (LReID).

PAD introduces a frozen text encoder as a stable semantic anchor across domains, and combines:

• TA-Prompt for text-side semantic alignment
• VA-Prompt for visual-side incremental adaptation
• Fixed textual distillation to preserve semantic consistency
• EMA-based visual distillation to mitigate semantic drift and catastrophic forgetting

The proposed method achieves strong performance on both seen and unseen domains under standard lifelong ReID benchmarks.

Status

Code will be released soon.

Paper

Prompt-Anchored Vision–Text Distillation for Lifelong Person Re-identification CVPR 2026

Contact

Wen Wen (zuzi666666@gmail.com)

Citation

@inproceedings{wen2026pad,
  title={Prompt-Anchored Vision–Text Distillation for Lifelong Person Re-identification},
  author={Wen, Wen and Chen, Hao and Zhang, Shiliang},
  booktitle={CVPR},
  year={2026}
}