Research map / Label-efficient and robust learning / Semi-supervised learning
Transductive Centroid Projection for Semi-supervised Large-scale Recognition
Yu Liu, Guanglu Song, Jing Shao, Xiao Jin, Xiaogang 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 |
|---|---|---|---|
| add self-training on clustered unlabeled data (ablation) | SU (cluster IDs only) · Market-1501 | Top-1 accuracy: 22.8 -> 65.0 | Improved |
| add self-training to the labeled+unlabeled combination (ablation) | MU+L · Market-1501 | Top-1 accuracy: 37.4 -> 68.8 | Improved |
| replace naive combination with TCP on unlabeled data (ablation) | MU+L · Market-1501 | Top-1 accuracy: 37.4 -> 89.6 | Improved |
| use real annotated ground truth instead of pseudo cluster IDs for unlabeled data (ablation) | SU · Market-1501 | Top-1 accuracy: 22.8 -> 66.4 | Improved |
| add manually annotated unlabeled data to labeled training (ablation) | SL · Market-1501 | Top-1 accuracy: 87.7 -> 86.0 | Hurt |
| add naive unlabeled cluster data to labeled data (ablation) | SL · Market-1501 | Top-1 accuracy: 87.7 -> 37.4 | Hurt |
| add triplet loss on unlabeled data (ablation) | MU+L · Market-1501 | Top-1 accuracy: 37.4 -> 83.5 | Improved |
| use TCP instead of triplet loss on unlabeled data (ablation) | MU+L triplet loss · Market-1501 | Top-1 accuracy: 83.5 -> 89.6 | Improved |
| use TCP instead of self-training (ablation) | MU+L self-training · Market-1501 | Top-1 accuracy: 68.8 -> 89.6 | Improved |
| add TCP semi-supervised training to the labeled baseline (ablation) | SL · Market-1501 | Top-1 accuracy: 87.7 -> 90.4 | Improved |
| add re-ranking to TCP | TCP · Market-1501 | mAP: 66.2 -> 84.4 | Improved |
| add TCP unlabeled-data training to baseline | Baseline · CUHK03 | Top-1 accuracy: 91.7 -> 94.4 | Improved |
Closest papers by meaning
- Distant Supervised Centroid Shift: A Simple and Efficient Approach to Visual Domain Adaptation (CVPR 2019 · similarity 0.84)
- Transductive Semi-Supervised Deep Learning using Min-Max Features (ECCV 2018 · similarity 0.83)
- Semi-Supervised Learning Made Simple With Self-Supervised Clustering (CVPR 2023 · similarity 0.83)
- Label Propagation for Deep Semi-Supervised Learning (CVPR 2019 · similarity 0.82)
- Revisiting Semi-Supervised Learning in the Era of Foundation Models (NeurIPS 2025 · similarity 0.81)
- Targeted Representation Alignment for Open-World Semi-Supervised Learning (CVPR 2024 · similarity 0.81)
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.
