Research map / Label-efficient and robust learning / Test-time adaptation
Generalizable Person Re-identification via Self-Supervised Batch Norm Test-Time Adaption
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 |
|---|---|---|---|
| Add test-time BN adaptation using the two part-based self-supervised losses (ablation) | BNTA · VIPeR | rank-1: 4.9% | Helped |
| Add test-time BN adaptation using the two part-based self-supervised losses (ablation) | BNTA · GRID | rank-1: 8.7% | Helped |
| Add test-time BN adaptation using the two part-based self-supervised losses (ablation) | BNTA · iLIDS | rank-1: 1.7% | Helped |
| Add joint part-positioning and part-matching SSL training to identity training (ablation) | BNTA · VIPeR | rank-1: 50.9 -> 52.8 | Helped |
| Use both part-positioning and part-matching losses during TTA (ablation) | BNTA · VIPeR | rank-1: 56.8 -> 57.4 | Helped |
| Update BN parameters rather than convolution parameters during TTA (ablation) | BNTA · GRID | rank-1: 47.8 -> 51.1 | Helped |
| Add IN updates to BN updates during TTA (ablation) | BNTA · GRID | rank-1: 51.1 -> 50.9 | Hurt |
| Add convolution updates to BN updates during TTA (ablation) | BNTA · GRID | rank-1: 51.1 -> 48.1 | Hurt |
| Update all BN statistics and affine parameters rather than statistics alone (ablation) | BNTA · GRID | rank-1: 45.2 -> 51.1 | Helped |
| Increase the number of target samples used for adaptation while holding K fixed (ablation) | BNTA · GRID | rank-1 | Helped |
Closest papers by meaning
- Meta Batch-Instance Normalization for Generalizable Person Re-Identification (CVPR 2021 · similarity 0.93)
- Person30K: A Dual-Meta Generalization Network for Person Re-Identification (CVPR 2021 · similarity 0.90)
- Style Normalization and Restitution for Generalizable Person Re-Identification (CVPR 2020 · similarity 0.90)
- Unsupervised Multi-Source Domain Adaptation for Person Re-Identification (CVPR 2021 · similarity 0.89)
- Generalizable Person Re-identification via Balancing Alignment and Uniformity (NeurIPS 2024 · similarity 0.89)
- Exploring the Camera Bias of Person Re-identification (ICLR 2025 · similarity 0.87)
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.