Research map / Label-efficient and robust learning / Domain adaptation and generalization
Dynamically Transformed Instance Normalization Network for Generalizable Person Re-identification
Bingliang Jiao, Lingqiao Liu, Liying Gao, Guosheng Lin, Lu Yang, Shizhou Zhang, Peng Wang, Yanning Zhang
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 |
|---|---|---|---|
| add DTIN module to ResNet-50 baseline with IN (ablation) | Base+IN · GRID (Protocol-1) | mAP: 52.7 -> 56.4 | Improved |
| add DTIN module on VIPeR (ablation) | Base+IN · VIPeR (Protocol-1) | mAP: 67.5 -> 69.1 | Improved |
| add multi-task training to DTIN-Net (ablation) | DTIN-Net · GRID (Protocol-1) | mAP: 56.4 -> 60.6 | Improved |
| add multi-task training to DTIN-Net (ablation) | DTIN-Net · VIPeR (Protocol-1) | mAP: 69.1 -> 70.7 | Improved |
| add instance normalization to baseline (ablation) | Baseline · GRID (Protocol-1) | mAP: 46.5 -> 52.7 | Improved |
| add dynamic control path (ablation) | DTIN-Net with multi-task training · GRID (Protocol-1) | CMC-1: 49.3 -> 51.8 | Improved |
| add dynamic control path (ablation) | DTIN-Net with multi-task training · VIPeR (Protocol-1) | CMC-1: 62.3 -> 62.9 | Improved |
| add dynamic control path (ablation) | DTIN-Net without multi-task training · GRID (Protocol-1) | mAP: 54.7 -> 56.4 | Improved |
| concatenate IN with dynamic convolution instead of using DTIN (ablation) | Base+IN+DyConv · GRID/VIPeR (Protocol-1) | CMC | No clear effect |
| generate dynamic filters from normalized instead of unnormalized features (ablation) | DTIN · GRID/VIPeR (Protocol-1) | CMC | Hurt |
| stack all domains with a single classifier and triplet loss instead of multi-task (ablation) | DTIN-Net · Protocol-1 | accuracy | Hurt |
| DTIN-Net vs Base+IN capacity | DTIN-Net · GRID (Protocol-1) | mAP: 8.3% mAP gain with 2 M extra parameters and 0.4 GMac fewer | Improved |
Closest papers by meaning
- Rethinking Normalization Layers for Domain Generalizable Person Re-identification (ECCV 2024 · similarity 0.92)
- Meta Batch-Instance Normalization for Generalizable Person Re-Identification (CVPR 2021 · similarity 0.92)
- Style Normalization and Restitution for Generalizable Person Re-Identification (CVPR 2020 · similarity 0.91)
- Person30K: A Dual-Meta Generalization Network for Person Re-Identification (CVPR 2021 · similarity 0.91)
- Generalizable Person Re-identification via Self-Supervised Batch Norm Test-Time Adaption (AAAI 2022 · similarity 0.90)
- Generalizable Person Re-identification via Balancing Alignment and Uniformity (NeurIPS 2024 · similarity 0.89)
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.
