Research map / Label-efficient and robust learning / Recognition and long-tail
Structured Semantic Transfer for Multi-Label Recognition with Partial Labels
Read the paperCheck 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 intra-image and cross-image semantic transfer to generate pseudo labels | SSGRL on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 74.1 -> 76.7; improvement of 2.6% | Helped |
| Replace dataset-level statistical co-occurrence with image-specific co-occurrence (ablation) | IST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 69.1 -> 75.2; 6.1% | Helped |
| Add the IST loss (ablation) | IST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 74.3 -> 75.2; 0.9% | Helped |
| Add the CST loss (ablation) | CST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 74.5 -> 75.4 | Helped |
| Use IST only rather than the full SST framework (ablation) | SST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 76.7 -> 75.2 | Hurt |
| Use CST only rather than the full SST framework (ablation) | SST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 76.7 -> 75.4 | Hurt |
| Replace semantic decoupling with SAM (ablation) | SST on MS-COCO · MS-COCO, 10%-90% known labels | average mAP: 76.7 -> 76.5 | Hurt |
| Decrease the minimum intra-image threshold from 0.8 to 0.75 (ablation) | IST on MS-COCO · MS-COCO, 20% and 50% known labels | mAP | Helped |
| Decrease the minimum intra-image threshold below 0.75 (ablation) | IST on MS-COCO · MS-COCO, 20% and 50% known labels | mAP | Hurt |
| Decrease the minimum cross-image threshold from 0.8 to 0.75 (ablation) | CST on MS-COCO · MS-COCO, 20% known labels | mAP: 71.2% -> 72.5% | Helped |
| Decrease the minimum cross-image threshold from 0.8 to 0.75 (ablation) | CST on MS-COCO · MS-COCO, 50% known labels | mAP: 76.8% -> 77.1% | Helped |
| Decrease the minimum cross-image threshold below 0.75 (ablation) | CST on MS-COCO · MS-COCO, 20% and 50% known labels | mAP | Hurt |
Closest papers by meaning
- Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels (AAAI 2022 · similarity 0.91)
- Language-Driven Cross-Modal Classifier for Zero-Shot Multi-Label Image Recognition (ICML 2024 · similarity 0.86)
- Multi-Label Supervised Contrastive Learning (AAAI 2024 · similarity 0.86)
- Recover and Match: Open-Vocabulary Multi-Label Recognition through Knowledge-Constrained Optimal Transport (CVPR 2025 · similarity 0.86)
- In Pursuit of Causal Label Correlations for Multi-label Image Recognition (NeurIPS 2024 · similarity 0.85)
- Correlative and Discriminative Label Grouping for Multi-Label Visual Prompt Tuning (CVPR 2025 · similarity 0.85)
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.