Research map / Label-efficient and robust learning / Recognition and long-tail
Learning Invariance From Generated Variance for Unsupervised Person Re-Identification
Hao Chen, Yaohui Wang, Benoit Lagadec, Antitza Dantcheva, François Brémond
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 D-Mixup on top of mesh rotation (ablation) | GCL+ · Market-1501 (fully unsupervised) | mAP: 63.8 -> 66.3 | Improved |
| replace multiple traditional augmentations with mesh-guided rotation (ablation) | GCL+ contrastive module · Market-1501 (fully unsupervised) | mAP: 58.2 -> 63.8 | Improved |
| use D-Mixup instead of image-level Mixup (ablation) | GCL+ · Market-1501 (fully unsupervised) | mAP: 65.9 -> 66.3 | Improved |
| use D-Mixup instead of feature-level Mixup (ablation) | GCL+ · Market-1501 (fully unsupervised) | mAP: 66.1 -> 66.3 | Improved |
| add traditional multi augmentation in contrastive module (ablation) | JVTC baseline · Market-1501 (fully unsupervised) | mAP: 47.2 -> 58.2 | Improved |
| add view-invariant loss L_vi' (ablation) | GCL+ · Duke->Market (UDA) | mAP: 61.6 -> 69.1 | Improved |
| add view-invariant loss L_vi'' (ablation) | GCL+ · Duke->Market (UDA) | mAP: 69.1 -> 72.5 | Improved |
| add prototype loss L_proto (ablation) | GCL+ · Duke->Market (UDA) | mAP: 72.5 -> 72.8 | Improved |
| add mixed prototype loss L_mix (ablation) | GCL+ · Duke->Market (UDA) | mAP: 72.8 -> 74.4 | Improved |
| rotate one mesh instead of switching 2D gray images (ablation) | GCL+ · Duke->Market (UDA) | mAP: 60.1 -> 74.4 | Improved |
| rotate one mesh instead of switching two meshes (ablation) | GCL+ · Duke->Market (UDA) | mAP: 74.2 -> 74.4 | No clear effect |
| raise DBSCAN distance threshold from 0.5 to 0.6 (ablation) | GCL+ · Duke->Market (UDA) | mAP: 74.4 -> 72.4 | Hurt |
Closest papers by meaning
- Joint Generative and Contrastive Learning for Unsupervised Person Re-Identification (CVPR 2021 · similarity 0.95)
- Learning Disentangled Representation for Robust Person Re-identification (NeurIPS 2019 · similarity 0.92)
- Disentangled Representations for Short-Term and Long-Term Person Re-Identification (TPAMI 2021 · similarity 0.91)
- FD-GAN: Pose-guided Feature Distilling GAN for Robust Person Re-identification (NeurIPS 2018 · similarity 0.91)
- Pose-Normalized Image Generation for Person Re-identification (ECCV 2018 · similarity 0.90)
- From Poses to Identity: Training-Free Person Re-Identification via Feature Centralization (CVPR 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.
