Research map / Continual learning / Domain and prompt-based
Generalising without Forgetting for Lifelong Person Re-Identification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 classification, distribution, and representation coherence losses together (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 63.7 -> 71.1 | Helped |
| Add classification coherence instead of cross-entropy (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 64.7 -> 71.1 | Helped |
| Add distribution coherence loss (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 70.8 -> 71.1 | Helped |
| Add representation coherence loss (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 69.2 -> 71.1 | Helped |
| Add distribution and representation coherence together (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 65.1 -> 71.1 | Helped |
| Coordinate loss weights over training instead of fixing ω(e) at zero for coherence losses (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 65.1 -> 71.1 | Helped |
| Coordinate classification-loss weighting rather than using the rebalanced factor throughout training (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 69.7 -> 71.1 | Helped |
| Use constant distribution and representation coherence losses throughout training (ablation) | GwFReID · Four sequential Re-ID benchmarks | Rank-1 accuracy: 71.1 -> 65.4 | Hurt |
| Set representation-coherence weight λ=0 (ablation) | GwFReID · Four sequential Re-ID benchmarks | forgetting ratio: 6.7 -> 9.9 | Hurt |
| Align all samples to the old feature space rather than limiting alignment to exemplars (ablation) | GwFReID · Four sequential Re-ID benchmarks | Average incremental accuracy: 78.0 -> 74.9 | Hurt |
| Increase the exemplar memory from K=2 to K=4 (ablation) | GwFReID · Four sequential Re-ID benchmarks | Average incremental accuracy: 78.0 -> 79.2 | Helped |
| Use herding rather than random exemplar selection (ablation) | GwFReID · Four sequential Re-ID benchmarks | Average incremental Rank-1 accuracy: 77.4 -> 78.0 | No clear effect |
Closest papers by meaning
- Towards Better Generalization in Lifelong Person Re-Identification with Flatness-Aware Learning (NeurIPS 2026 · similarity 0.90)
- Lifelong Person Re-identification by Pseudo Task Knowledge Preservation (AAAI 2022 · similarity 0.90)
- Lifelong Person Re-Identification via Adaptive Knowledge Accumulation (CVPR 2021 · similarity 0.90)
- Distribution-aware Knowledge Prototyping for Non-exemplar Lifelong Person Re-identification (CVPR 2024 · similarity 0.89)
- Generalizable Person Re-identification via Balancing Alignment and Uniformity (NeurIPS 2024 · similarity 0.88)
- 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.