Research map / Label-efficient and robust learning / Noisy labels
Large Loss Matters in Weakly Supervised Multi-Label Classification
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 |
|---|---|---|---|
| Apply a relative large-loss criterion rather than treating all unobserved labels as negative | Naive AN · VOC, COCO, NUSWIDE, and CUB | mAP: 85.1 -> 89.2; 64.1 -> 71.0; 42.0 -> 48.3; 19.1 -> 20.4 | Helped |
| Apply a relative large-loss criterion | Naive AN · OpenImages V3 (All Gs) | mAP: 82.0 -> 82.6 | Helped |
| Replace the relative threshold with an absolute threshold in LL-R (ablation) | LL-relative · VOC, COCO, NUSWIDE, and CUB | mAP: 89.2 -> 89.0; 71.0 -> 70.3; 48.3 -> 47.4; 20.4 -> 20.1 | Mixed |
| Use an absolute threshold with LL-Ct (ablation) | LL-relative · VOC, COCO, NUSWIDE, and CUB | mAP: 89.2 -> 89.0; 71.0 -> 70.3; 48.3 -> 48.6; 20.4 -> 19.8 | Mixed |
| Use an absolute threshold with LL-Cp (ablation) | LL-relative · VOC, COCO, NUSWIDE, and CUB | mAP: 89.2 -> 88.7; 71.0 -> 70.5; 48.3 -> 48.0; 20.4 -> 19.8 | Mixed |
| Replace the relative threshold with an absolute threshold in LL-R (ablation) | LL-relative · OpenImages V3 (All Gs) | mAP: 82.6 -> 82.3 | Hurt |
| Use an absolute threshold with LL-Ct (ablation) | LL-relative · OpenImages V3 (All Gs) | mAP: 82.6 -> 82.4 | Hurt |
| Use an absolute threshold with LL-Cp (ablation) | LL-relative · OpenImages V3 (All Gs) | mAP: 82.6 -> 82.3 | Hurt |
Closest papers by meaning
- Losses over Labels: Weakly Supervised Learning via Direct Loss Construction (AAAI 2023 · similarity 0.90)
- Multi-Label Classification With Partial Annotations Using Class-Aware Selective Loss (CVPR 2022 · similarity 0.90)
- Weakly Supervised Image Classification Through Noise Regularization (CVPR 2019 · similarity 0.89)
- Bridging the Gap Between Model Explanations in Partially Annotated Multi-Label Classification (CVPR 2023 · similarity 0.89)
- Exploiting weakly supervised visual patterns to learn from partial annotations (NeurIPS 2020 · similarity 0.88)
- Revisiting Interpolation for Noisy Label Correction (AAAI 2025 · 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.