Research map / Video / Video segmentation and tracking
CycAs: Self-supervised Cycle Association for Learning Re-identifiable Descriptions
Zhongdao Wang, Jingwei Zhang, Liang Zheng, Yixuan Liu, Yifan Sun, Yali Li, Shengjin Wang
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 |
|---|---|---|---|
| use relaxed margin-based asymmetric loss instead of symmetric L1 cycle loss on symmetric data (ablation) | CycAs · Market-1501 symmetric data | rank-1: 78.9 -> 88.1 | Improved |
| use relaxed margin-based asymmetric loss on asymmetric data (ablation) | CycAs · Market-1501 asymmetric data | rank-1: 67.1 -> 84.8 | Improved |
| use relaxed margin-based asymmetric loss on asymmetric data (ablation) | CycAs · Market-1501 asymmetric data | mAP: 47.1 -> 64.8 | Improved |
| add Stage I intra-sampling training before Stage II (ablation) | CycAs · Market-1501 | rank-1: 84.6 -> 84.8 | No clear effect |
| add inter-sampled Stage II after intra-sampled Stage I (ablation) | CycAs · Market-1501 | rank-1: 29.4 -> 84.8 | Improved |
| include intra-sampled data in Stage II training (ablation) | CycAs · Market-1501 | rank-1: - -> 84.6 | Improved |
| lengthen intra-sampling interval from 2 s to 4 s (Stage I + II) (ablation) | CycAs · Market-1501 | rank-1: 84.8 -> 84.2 | No clear effect |
| lengthen intra-sampling interval from 2 s to 8 s (Stage I + II) (ablation) | CycAs · Market-1501 | mAP: 64.8 -> 53.3 | Hurt |
| lengthen intra-sampling interval from 2 s to 4 s (Stage I only) (ablation) | CycAs · Market-1501 | rank-1: 29.4 -> 25.8 | Hurt |
| raise inter-sampling symmetry tau_beta from 0.6 toward 1 (ablation) | CycAs · Market-1501 | rank-1: 84.8 -> 88.1 | Improved |
| vary intra-sampling symmetry tau_alpha from 1 to 0.3 (ablation) | CycAs · Market-1501 | rank-1: 84.2 -> 84.2 | No clear effect |
| replace supervised cross-domain IDE with CycAs trained on unlabeled videos | CycAs · Market-1501 (self-collected video) | rank-1: 48.1 -> 50.8 | Improved |
Closest papers by meaning
- Tracklet Self-Supervised Learning for Unsupervised Person Re-Identification (AAAI 2020 · similarity 0.88)
- Unsupervised Graph Association for Person Re-Identification (ICCV 2019 · similarity 0.88)
- Distilled Person Re-Identification: Towards a More Scalable System (CVPR 2019 · similarity 0.88)
- Self-Similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-Identification (ICCV 2019 · similarity 0.88)
- Weakly Supervised Tracklet Association Learning With Video Labels for Person Re-Identification (TPAMI 2023 · similarity 0.88)
- Learning from Synchronization: Self-Supervised Uncalibrated Multi-View Person Association in Challenging Scenes (CVPR 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.
