Research map / Label-efficient and robust learning / Recognition and long-tail
ICE: Inter-Instance Contrastive Encoding for Unsupervised Person Re-Identification
Hao Chen, Benoit Lagadec, François Bremond
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 hard instance contrastive loss (ablation) | camera-aware proxy baseline · Market1501 | mAP: 79.3 -> 80.5 | Improved |
| add soft instance consistency loss (ablation) | camera-aware proxy baseline · Market1501 | mAP: 79.3 -> 81.1 | Improved |
| add both losses (ablation) | camera-aware proxy baseline · MSMT17 | mAP: 36.4 -> 38.9 | Improved |
| add both losses (ablation) | camera-aware proxy baseline · DukeMTMC-reID | mAP: 67.3 -> 69.9 | Improved |
| add hard instance contrastive loss (ablation) | camera-agnostic baseline · Market1501 | mAP: 65.8 -> 78.2 | Improved |
| add soft instance consistency loss alone (ablation) | camera-agnostic baseline · Market1501 | mAP: 65.8 -> 47.2 | Hurt |
| add soft instance consistency loss on top of hard loss (ablation) | camera-agnostic ICE · Market1501 | mAP: 78.2 -> 79.5 | Improved |
| add soft instance consistency loss on top of hard loss (ablation) | camera-agnostic ICE · MSMT17 | mAP: 30.3 -> 29.8 | Hurt |
| use all negatives instead of hardest negative in hard loss (ablation) | ICE (aware) · Market1501 | mAP: 80.1 -> 82.3 | Improved |
| use KL with unaugmented target instead of MSE (ablation) | ICE (aware) · Market1501 | mAP: 80.0 -> 82.3 | Improved |
| use KL with unaugmented target instead of strong-strong augmentation (ablation) | ICE (aware) · Market1501 | mAP: 80.4 -> 82.3 | Improved |
| IBN-ResNet50 backbone instead of ResNet50 | ICE (aware) · MSMT17 | mAP: 38.9 -> 40.6 | Improved |
Closest papers by meaning
- Unleashing Potential of Unsupervised Pre-Training With Intra-Identity Regularization for Person Re-Identification (CVPR 2022 · similarity 0.90)
- Discrepant and Multi-Instance Proxies for Unsupervised Person Re-Identification (ICCV 2023 · similarity 0.89)
- Unsupervised Person Re-Identification by Soft Multilabel Learning (CVPR 2019 · similarity 0.89)
- Part-Based Pseudo Label Refinement for Unsupervised Person Re-Identification (CVPR 2022 · similarity 0.89)
- Intra-Inter Domain Similarity for Unsupervised Person Re-Identification (TPAMI 2022 · similarity 0.89)
- Learning Commonality, Divergence and Variety for Unsupervised Visible-Infrared Person Re-identification (NeurIPS 2024 · 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.
