Research map / Label-efficient and robust learning / Recognition and long-tail
Acknowledging the Unknown for Multi-Label Learning with Single Positive Labels
Donghao Zhou, Pengfei Chen, Qiong Wang, Guangyong Chen, Pheng-Ann Heng
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 |
|---|---|---|---|
| replace assuming-negative loss with entropy-maximization loss on unannotated labels | ResNet-50 SPML · CUB | mAP: 18.31 -> 20.85 | Improved |
| replace assuming-negative loss with entropy-maximization loss | ResNet-50 SPML · COCO | mAP: 64.92 -> 70.70 | Improved |
| replace assuming-negative loss with entropy-maximization loss | ResNet-50 SPML · VOC | mAP: 85.89 -> 89.09 | Improved |
| add asymmetric pseudo-labeling to EM loss | EM loss · CUB | mAP: 20.85 -> 21.84 | Improved |
| add asymmetric pseudo-labeling to EM loss | EM loss · VOC | mAP: 89.09 -> 89.19 | Improved |
| add entropy minimization regularization instead of entropy maximization | AN loss · VOC | mAP: 85.89 -> 53.16 | Hurt |
| use high sample proportion (90%) for negative pseudo-labels instead of 10% (ablation) | APL · CUB | mAP: 20.95 -> 21.84 | Improved |
| use soft pseudo-labels instead of hard ones (ablation) | APL · CUB | mAP: 20.99 -> 21.84 | Improved |
| add down-weighting of pseudo-label loss (beta<1) (ablation) | APL · CUB | mAP: 21.00 -> 21.84 | Improved |
| add positive pseudo-labeling at 10% proportion (ablation) | APL · CUB | mAP: 21.84 -> 20.82 | Hurt |
| use ResNet-101 backbone instead of ResNet-50 (ablation) | EM loss · VOC | mAP: 89.09 -> 88.80 | Hurt |
| use EM loss instead of AN loss for APL pseudo-label generation (ablation) | APL · VOC | pseudo-label precision (%): 96.44 -> 99.57 | Improved |
Closest papers by meaning
- More Reliable Pseudo-labels, Better Performance: A Generalized Approach to Single Positive Multi-label Learning (ICCV 2025 · similarity 0.92)
- Revisiting Pseudo-Label for Single-Positive Multi-Label Learning (ICML 2023 · similarity 0.91)
- Can Class-Priors Help Single-Positive Multi-Label Learning? (NeurIPS 2025 · similarity 0.89)
- Learning with Partial-Label and Unlabeled Data: A Uniform Treatment for Supervision Redundancy and Insufficiency (ICML 2024 · similarity 0.89)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.89)
- Learning with Complementary Labels Revisited: The Selected-Completely-at-Random Setting Is More Practical (ICML 2024 · 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.
