Research map / Label-efficient and robust learning / Recognition and long-tail
Multi-Label Learning From Single Positive Labels
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 ROLE to estimate unobserved labels online under an expected-positive constraint instead of AN-LS | Single-positive linear classifier · VOC12 test | MAP: 85.3 -> 86.5 | Helped |
| Use ROLE to estimate unobserved labels online under an expected-positive constraint instead of AN-LS | Single-positive linear classifier · COCO test | MAP: 64.8 -> 66.3 | Helped |
| Use ROLE to estimate unobserved labels online under an expected-positive constraint instead of AN-LS | Single-positive linear classifier · NUS test | MAP: 48.5 -> 49.5 | Helped |
| Use ROLE to estimate unobserved labels online under an expected-positive constraint instead of AN-LS | Single-positive linear classifier · CUB test | MAP: 15.4 -> 16.2 | Hurt |
| Add label smoothing to assume-negative training (ablation) | Single-positive multi-label classifier · Most reported test settings | MAP | Helped |
| Add expected-positive regularization alone (ablation) | Single-positive classifier · VOC12, COCO, NUS test | MAP | Hurt |
| Add expected-positive regularization alone (ablation) | Single-positive classifier · CUB test | MAP: 17.2 -> 18.0 (linear); 19.1 -> 20.0 (fine-tuned) | Helped |
| Add online label estimation and expected-positive constraints together (ROLE) rather than expected-positive regularization alone (ablation) | Single-positive classifier · Linear test: VOC12, COCO, NUS; fine-tuned test: VOC12 | MAP: 83.8 -> 86.5; 62.6 -> 66.3; 46.4 -> 49.5; 85.5 -> 87.9 | Helped |
| Add linear-training initialization before end-to-end fine-tuning (ablation) | ROLE · VOC12, COCO, NUS, CUB test | MAP: 87.9 -> 88.2; 66.3 -> 69.0; 43.1 -> 51.0; 15.0 -> 16.8 | Mixed |
| Add linear-training initialization before end-to-end fine-tuning (ablation) | AN-LS · VOC12, COCO, NUS, CUB test | MAP: 86.7 -> 86.5; 66.9 -> 69.2; 44.9 -> 50.5; 17.9 -> 16.6 | Mixed |
| Apply negative-only asymmetric label smoothing instead of symmetric AN-LS (ablation) | Single-positive fine-tuned classifier · VOC12 and COCO test | MAP: +1.0 MAP for VOC12, +0.7 MAP for COCO | Helped |
| Apply negative-only asymmetric label smoothing instead of symmetric AN-LS (ablation) | Single-positive linear classifier · VOC12 and COCO test | MAP: -0.3 MAP for VOC12, -0.4 MAP for COCO | Hurt |
Closest papers by meaning
- Learning a Deep ConvNet for Multi-Label Classification With Partial Labels (CVPR 2019 · similarity 0.91)
- Bridging the Gap Between Model Explanations in Partially Annotated Multi-Label Classification (CVPR 2023 · similarity 0.90)
- Re-Labeling ImageNet: From Single to Multi-Labels, From Global to Localized Labels (CVPR 2021 · similarity 0.90)
- Noise-Robust Learning from Multiple Unsupervised Sources of Inferred Labels (AAAI 2022 · similarity 0.90)
- Addressing Multi-Label Learning with Partial Labels: From Sample Selection to Label Selection (AAAI 2025 · similarity 0.89)
- Multi-Label Image Recognition With Graph Convolutional Networks (CVPR 2019 · similarity 0.89)
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.