Research map / Label-efficient and robust learning / Recognition and long-tail
Learning a Deep ConvNet for Multi-Label Classification With Partial 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 partial-BCE instead of standard BCE | ResNet-101 WELDON on Open Images · Open Images | MAP (%): 79.01 -> 83.05 | Helped |
| Add GNN to the partial-BCE model (ablation) | ResNet-101 WELDON on Open Images · Open Images | MAP (%): 83.05 -> 83.36 | Helped |
| Use 10% clean partial labels instead of noisy+ labels | MS COCO classifier · MS COCO | MAP (%): 71.60 -> 72.15 | Helped |
| Label all images partially rather than fully labeling a subset | ResNet-101 WELDON · MS COCO val2014, 10% labels | MAP: 65.22 -> 72.15 | Helped |
| Fine-tune the model (ablation) | MS COCO, 10% known labels · MS COCO | MAP: 66.21 -> 72.15 | Helped |
| Replace BCE with partial-BCE (ablation) | Fine-tuned model · MS COCO, 10% known labels | MAP: 72.15 -> 75.31 | Helped |
| Add a GNN to partial-BCE (ablation) | Fine-tuned partial-label classifier · MS COCO, 10% known labels | MAP: 75.31 -> 75.82 | Helped |
| Add relabeling to partial-BCE (ablation) | Fine-tuned partial-label classifier · MS COCO, 10% known labels | 0-1 exact match: 24.51 -> 30.52 | Helped |
| Predict all missing labels at once instead of using curriculum relabeling (ablation) | Partial-label classifier · Pascal VOC 2007 val | MAP relative score: -1.49 | Hurt |
| Use Bayesian uncertainty relabeling rather than score-threshold relabeling (ablation) | Partial-label classifier with GNN · Pascal VOC 2007 val | MAP relative score: 0.34 -> 0.59 | Helped |
| Use ensemble scores instead of a single model score (ablation) | Score-threshold curriculum relabeling · Pascal VOC 2007 val | MAP relative score: 0.34 -> 0.23 | No clear effect |
| Predict only positive missing labels (ablation) | Curriculum relabeling · Pascal VOC 2007 val | 0-1 exact match relative score: -4.58 | Hurt |
Closest papers by meaning
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.91)
- Interactive Multi-Label CNN Learning With Partial Labels (CVPR 2020 · similarity 0.89)
- CNN-RNN: A Unified Framework for Multi-Label Image Classification (CVPR 2016 · similarity 0.89)
- Multi-Label Classification With Partial Annotations Using Class-Aware Selective Loss (CVPR 2022 · similarity 0.89)
- Realistic Evaluation of Deep Partial-Label Learning Algorithms (ICLR 2025 · similarity 0.86)
- Incomplete Multi-View Multi-label Learning via Disentangled Representation and Label Semantic Embedding (CVPR 2025 · similarity 0.86)
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.