Research map / Label-efficient and robust learning / Domain adaptation and generalization
Self-paced Contrastive Learning with Hybrid Memory for Domain Adaptive Object Re-ID
Yixiao Ge, Feng Zhu, Dapeng Chen, Rui Zhao, hongsheng Li
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 hybrid source-class, target-cluster, and target-instance supervision with self-paced clustering (ablation) | SpCL · MSMT17→Market-1501 | mAP: 69.2 -> 77.5 | Improved |
| Add self-paced learning to target cluster training (ablation) | Source classes + target clusters · MSMT17→Market-1501 | mAP: 28.9 -> 69.2 | Improved |
| Use each target instance as a distinct class rather than clustering instances (ablation) | UDA training with source classes · MSMT17→Market-1501 | mAP: 25.3 -> 4.9 | Hurt |
| Include target-domain supervision alongside source classes (ablation) | Source-class-only training · MSMT17→Market-1501 | mAP: 25.3 -> 77.5 | Improved |
| Use source class centroids instead of learnable source weights (ablation) | SpCL · MSMT17→Market-1501 | mAP: 77.5 -> 70.7 | Hurt |
| Remove unified contrast across supervision types (ablation) | SpCL · MSMT17→Market-1501 | mAP: 77.5 -> 68.8 | Hurt |
| Use independence and compactness reliability criteria (ablation) | SpCL self-paced clustering · MSMT17→Market-1501 | mAP: 74.5 -> 77.5 | Improved |
| Use independence and compactness reliability criteria (ablation) | SpCL without source data · Market-1501 | mAP: 68.2 -> 73.1 | Improved |
| Apply self-paced learning to agglomerative clustering (ablation) | Unsupervised SpCL · Market-1501 | mAP: 70.4 -> 75.2 | Improved |
| Replace the reliability criterion with HDBSCAN (ablation) | Unsupervised SpCL · Market-1501 | mAP: 73.1 -> 71.7 | Hurt |
| Replace ResNet-50 with IBN-ResNet (ablation) | SpCL · Person re-ID adaptation tasks | mAP | Mixed |
| Train jointly with unlabeled target data while retaining source-domain supervision | Source-domain encoder · MSMT17 source evaluation | mAP: 84.7 -> 86.8 | Improved |
Closest papers by meaning
- Multi-Centroid Representation Network for Domain Adaptive Person Re-ID (AAAI 2022 · similarity 0.88)
- Towards Discriminative Representation Learning for Unsupervised Person Re-Identification (ICCV 2021 · similarity 0.87)
- Towards Effective Instance Discrimination Contrastive Loss for Unsupervised Domain Adaptation (ICCV 2023 · similarity 0.87)
- Reliability Exploration with Self-Ensemble Learning for Domain Adaptive Person Re-identification (AAAI 2022 · similarity 0.87)
- Contrastive Test-Time Adaptation (CVPR 2022 · similarity 0.87)
- Alice Benchmarks: Connecting Real World Re-Identification with the Synthetic (ICLR 2024 · similarity 0.85)
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.
