Research map / Label-efficient and robust learning / Recognition and long-tail
Heterogeneous Semantic Transfer for Multi-label Recognition with Partial Labels
Tianshui Chen, Tao Pu, Lingbo Liu, Yukai Shi, Zhijing Yang, Liang Lin
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 IST module to SSGRL (ablation) | SSGRL · MS-COCO | average mAP: 74.1 -> 76.4 | Improved |
| add CST module to SSGRL (ablation) | SSGRL · VG-200 | average mAP: 39.7 -> 42.7 | Improved |
| add both IST and CST (full HST) (ablation) | SSGRL · MS-COCO | average mAP: 74.1 -> 77.9 | Improved |
| use semantic-correlation pseudo labels instead of confidence-based pseudo labels (ablation) | HST · VG-200 | average mAP: 40.8 -> 44.8 | Improved |
| learn image-specific co-occurrence instead of statistical co-occurrence (ablation) | IST · MS-COCO | average mAP: 70.7 -> 76.4 | Improved |
| add L_ist loss (ablation) | IST · VOC 2007 | average mAP: 89.0 -> 90.0 | Improved |
| add L_cst loss (ablation) | CST · MS-COCO | average mAP: 74.8 -> 76.5 | Improved |
| use instance-prototype similarities instead of instance-level similarities (ablation) | CST · MS-COCO | average mAP: 75.5 -> 76.5 | Improved |
| use SD instead of SA for SARL (ablation) | HST · MS-COCO | average mAP: 77.7 -> 77.9 | No clear effect |
| use dynamic threshold learning instead of fixed threshold 0.7 (ablation) | HST · MS-COCO | average mAP: 77.5 -> 77.9 | Improved |
| use dynamic threshold learning instead of fixed threshold 0.1 (ablation) | HST · VG-200 | average mAP: 40.8 -> 44.8 | Improved |
| increase number of prototypes K from 10 to 20 (ablation) | CST · MS-COCO | average mAP: K from 1 to 10 improves, K=20 decreases | Hurt |
Closest papers by meaning
- Structured Semantic Transfer for Multi-Label Recognition with Partial Labels (AAAI 2022 · similarity 0.96)
- Semantic-Aware Representation Blending for Multi-Label Image Recognition with Partial Labels (AAAI 2022 · similarity 0.92)
- Improving Multi-label Learning with Missing Labels by Structured Semantic Correlations (ECCV 2016 · similarity 0.89)
- Learning Semantic-Specific Graph Representation for Multi-Label Image Recognition (ICCV 2019 · similarity 0.89)
- DualCoOp++: Fast and Effective Adaptation to Multi-Label Recognition With Limited Annotations (TPAMI 2023 · similarity 0.88)
- [CLS] is Not Enough: Multi-Label Recognition via Patch-Level Inference and Adaptive Aggregation (ICML 2026 · 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.
