Research map / Label-efficient and robust learning / Recognition and long-tail
More Reliable Pseudo-labels, Better Performance: A Generalized Approach to Single Positive Multi-label Learning
Luong Tran, Thieu Vo, Anh Nguyen, Sang Dinh, Van Nguyen
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 GPR loss to random pseudo-labels (ablation) | BCE + Random pseudo-labels · Average over 4 datasets (SPML) | mAP: 51.56 -> 53.98 (+ 2.42) | Improved |
| add GPR loss to VLPL pseudo-labels (ablation) | VLPL · Average over 4 datasets (SPML) | mAP: 58.53 -> 59.13 (+ 0.6) | Improved |
| add GPR loss to LL-Ct pseudo-labels (ablation) | LL-Ct · Average over 4 datasets (SPML) | mAP: 56.98 -> 57.32 (+ 0.34) | Improved |
| add GPR loss to DAMP pseudo-labels (ablation) | BCE + DAMP · Average over 4 datasets (SPML) | mAP: 58.31 -> 59.90 (+ 1.59) | Improved |
| add GPR loss to LL-Ct on NUS-WIDE (ablation) | LL-Ct · NUS (SPML) | mAP: 48.00 -> 47.72 | Hurt |
| add noise addition G to DAMP (with T) (ablation) | AEVLP · VOC (SPML) | mAP: 90.28 -> 90.21 | Mixed |
| add positive pseudo-label re-weighting v4 on top of G, T, r (ablation) | AEVLP · CUB (SPML) | mAP: 24.24 -> 24.73 | Improved |
| add regularization term R (ablation) | AEVLP · VOC (SPML) | mAP: 90.31 -> 90.38 | Improved |
| add negative pseudo-label loss (ablation) | AEVLP · VOC (SPML) | mAP: 90.38 -> 90.46 | Improved |
| use grid size 4 instead of 2 (ablation) | AEVLP · VOC (SPML) | mAP: 90.13 -> 90.46 | Improved |
| use grid size 5 instead of 4 (ablation) | AEVLP · CUB (SPML) | mAP: 24.48 -> 24.89 | Improved |
| replace ResNet-50 with ConvNeXt-L-22K backbone (ablation) | AEVLP · COCO (SPML) | mAP: 73.54 -> 82.69 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- MIME — MiME: Multilevel Medical Embedding of Electronic Health Records for Predictive Healthcare (NeurIPS 2018) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Revisiting Pseudo-Label for Single-Positive Multi-Label Learning (ICML 2023 · similarity 0.92)
- Acknowledging the Unknown for Multi-Label Learning with Single Positive Labels (ECCV 2022 · similarity 0.92)
- Multi-Label Learning From Single Positive Labels (CVPR 2021 · similarity 0.91)
- SemPPL: Predicting Pseudo-Labels for Better Contrastive Representations (ICLR 2023 · similarity 0.89)
- DualCoOp++: Fast and Effective Adaptation to Multi-Label Recognition With Limited Annotations (TPAMI 2023 · similarity 0.89)
- Double-Layer Hybrid-Label Identification Feature Selection for Multi-View Multi-Label Learning (AAAI 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.
