Research map / Continual learning / Domain and prompt-based
AdalReID: Auto-selected Knowledge Adapters for Lifelong Person Re-identification
Xuelin Qian, Ruiqi Wu, Zhongling Huang, Dingwen Zhang, Junwei Han
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use temperature scheduling instead of one-hot similarity (fine-tuning) (ablation) | AdalReID · CUHK02 unseen | mAP: 79.6 -> 82.5 | Improved |
| use temperature scheduling instead of one-hot similarity (fine-tuning) (ablation) | AdalReID · seen-average | mAP: 84.2 -> 83.1 | Hurt |
| use temperature scheduling instead of uniform merging (t=2.0) (ablation) | AdalReID · seen-average | mAP: 78.2 -> 83.1 | Improved |
| use temperature scheduling instead of learnable t (ablation) | AdalReID · seen-average | mAP: 79.4 -> 83.1 | Improved |
| use 2-Wasserstein auto-selection instead of random selection (ablation) | AdalReID · seen-average | mAP: 69.7 -> 83.1 | Improved |
| use 2-Wasserstein auto-selection instead of latest selection (ablation) | AdalReID · seen-average | mAP: 71.4 -> 83.1 | Improved |
| use 2-Wasserstein instead of MMD for distribution distance (ablation) | AdalReID · CUHK02 unseen | Rank-1: 80.0 -> 81.4 | No clear effect |
| add Projection adapter on top of Q,V (ablation) | knowledge adapters · seen-average | mAP: 82.0 -> 82.9 | Improved |
| add FFN adapters to Q,K,V,Proj adapters (ablation) | knowledge adapters · seen-average | mAP: 83.1 -> 84.7 | Improved |
| use task-wise instead of sample-wise adapter selection (ablation) | AdalReID · Order-1 seen-average | mAP: 72.1 -> 76.5 | Improved |
| increase adapter rank from 16 to 64 (ablation) | knowledge adapters · seen domains | mAP | Improved |
| use large adapter scale alpha/r=8 (ablation) | knowledge adapters · seen domains | mAP | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LSTKC — LSTKC: Long Short-Term Knowledge Consolidation for Lifelong Person Re-identification (AAAI 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Lifelong Person Re-Identification via Adaptive Knowledge Accumulation (CVPR 2021 · similarity 0.93)
- Dynamic Magic: Unleashing Restricted Knowledge for Lifelong Person Re-Identification (CVPR 2026 · similarity 0.92)
- Distribution-aware Knowledge Prototyping for Non-exemplar Lifelong Person Re-identification (CVPR 2024 · similarity 0.90)
- LSTKC: Long Short-Term Knowledge Consolidation for Lifelong Person Re-identification (AAAI 2024 · similarity 0.90)
- DKC: Differentiated Knowledge Consolidation for Cloth-Hybrid Lifelong Person Re-identification (CVPR 2025 · similarity 0.89)
- DASK: Distribution Rehearsing via Adaptive Style Kernel Learning for Exemplar-Free Lifelong Person Re-Identification (AAAI 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
