Research map / Label-efficient and robust learning / Semi-supervised learning
Learning to Label: A Reinforced Self-Evolving Framework for Semi-supervised Referring Expression Segmentation
Runlong Cao, Ying Zang, Chuanwei Zhou, Tianrun Chen, Tong Zhang, Zhen Cui, Chunyan Xu
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 SPM uncertainty-aware prior fusion (ablation) | semi-supervised baseline · RefCOCO val 5% labels | overall IoU: 58.23 -> 65.73 | Improved |
| add SESM guidance on top of SPM (ablation) | L2L · RefCOCO val 5% labels | overall IoU: 65.73 -> 66.99 | Improved |
| add RPLE on top of SPM+SESM (ablation) | L2L · RefCOCO val 5% labels | overall IoU: 66.99 -> 67.30 | Improved |
| use learned stage-wise gating instead of fixed 0.5/0.5 weighting in SESM (ablation) | SESM · RefCOCO val 5% labels | overall IoU: 67.04 -> 67.30 | Improved |
| use projection in shallow stages and attention in deep stages instead of all attention (ablation) | SESM structural stream · RefCOCO val 5% labels | overall IoU: 65.08 -> 67.30 | Improved |
| use attention in all four stages instead of projection in all four (ablation) | SESM structural stream · RefCOCO val 5% labels | overall IoU: 66.92 -> 65.08 | Hurt |
| add consistency gain to RPLE reward (ablation) | RPLE · RefCOCO val 5% labels | overall IoU: 65.36 -> 67.30 | Improved |
| add coverage gain to RPLE reward (ablation) | RPLE · RefCOCO val 5% labels | overall IoU: 65.95 -> 67.30 | Improved |
| add separability gain to RPLE reward (ablation) | RPLE · RefCOCO val 5% labels | overall IoU: 66.35 -> 67.30 | Improved |
| replace RPLE adaptive thresholds with static foreground threshold (ablation) | L2L · RefCOCO val 5% labels | overall IoU: tau 0.7: 66.99; tau 0.8: 66.31 vs RPLE 67.30 | Mixed |
| increase SPM coefficients kappa_w or kappa_a (ablation) | SPM · RefCOCO val | overall IoU: 67.30 -> 65.96 at worst | No clear effect |
| use MLLM+SAM2 priors with naive 0.5 linear fusion vs no MLLM | semi-supervised baseline · RefCOCO val 0.1% labels | overall IoU: 21.72 -> 62.25 | Improved |
Closest papers by meaning
- SAM as the Guide: Mastering Pseudo-Label Refinement in Semi-Supervised Referring Expression Segmentation (ICML 2024 · similarity 0.91)
- Learning from Future: A Novel Self-Training Framework for Semantic Segmentation (NeurIPS 2022 · similarity 0.89)
- Learning To Segment Every Referring Object Point by Point (CVPR 2023 · similarity 0.89)
- SAFARI: Adaptive Sequence Transformer for Weakly Supervised Referring Expression Segmentation (ECCV 2024 · similarity 0.88)
- Weakly Supervised Referring Image Segmentation with Intra-Chunk and Inter-Chunk Consistency (ICCV 2023 · similarity 0.88)
- A Mutual Supervision Framework for Referring Expression Segmentation and Generation (IJCV 2025 · 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.
