Research map / Label-efficient and robust learning / Domain adaptation and generalization
Generalizing A Person Retrieval Model Hetero- and Homogeneously
Zhun Zhong, Liang Zheng, Shaozi Li, Yi Yang
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 triplet loss on source samples (ablation) | Baseline · Duke to Market-1501 | Rank-1 accuracy: 44.6 -> 48.6 | Improved |
| add camera invariance loss using camera-style-transferred target images (ablation) | Baseline · Duke to Market-1501 | Rank-1 accuracy: 44.6 -> 60.6 | Improved |
| add domain connectedness (unlabeled target negatives) on top of triplet loss (ablation) | Baseline + LT · Duke to Market-1501 | Rank-1 accuracy: 48.6 -> 49.8 | Improved |
| add domain connectedness (unlabeled target negatives) on top of triplet loss (ablation) | Baseline + LT · CUHK03 to Market-1501 | Rank-1 accuracy: 46.1 -> 48.9 | Improved |
| add joint camera invariance and domain connectedness over camera invariance alone (ablation) | Baseline + LC · CUHK03 to Market-1501 | Rank-1 accuracy: 53.6 -> 56.8 | Improved |
| add domain connectedness to camera invariance (ablation) | Baseline + LC · Duke to Market-1501 | mAP: 28.5 -> 31.4 | Improved |
| add domain connectedness to camera invariance (ablation) | Baseline + LC · Market-1501 to Duke | mAP: 22.1 -> 27.2 | Improved |
| add camera invariance loss (ablation) | Baseline · Market-1501 to Duke | Rank-1 accuracy: 32.9 -> 42.5 | Improved |
| add camera invariance loss (ablation) | Baseline · CUHK03 to Duke | Rank-1 accuracy: 24.3 -> 40.9 | Improved |
| add camera style transfer (vs crop or flip) as perturbation | Source-only baseline · Duke to Market-1501 | Mean feature distance: 0.049 -> 0.485 | Mixed |
| use cluster-based or supervised instead of random sampling of target images (ablation) | HHL · Duke/Market/CUHK03 transfers | Rank-1 accuracy and mAP | No clear effect |
| increase beta from 0 (baseline) to 0.4-0.8 (ablation) | HHL · four transfer settings | Rank-1 accuracy and mAP | Improved |
Closest papers by meaning
- Unsupervised Multi-Source Domain Adaptation for Person Re-Identification (CVPR 2021 · similarity 0.92)
- Camera-Driven Representation Learning for Unsupervised Domain Adaptive Person Re-identification (ICCV 2023 · similarity 0.92)
- A Novel Unsupervised Camera-Aware Domain Adaptation Framework for Person Re-Identification (ICCV 2019 · similarity 0.91)
- Towards Discriminative Representation Learning for Unsupervised Person Re-Identification (ICCV 2021 · similarity 0.91)
- Generalizing Person Re-Identification by Camera-Aware Invariance Learning and Cross-Domain Mixup (ECCV 2020 · similarity 0.91)
- Unsupervised Domain Adaptation for Person Re-identification via Heterogeneous Graph Alignment (AAAI 2021 · similarity 0.91)
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.
