Research map / Detection and segmentation / Open-set and few-shot detection
Refining Context-Entangled Content Segmentation via Curriculum Selection and Anti-Curriculum Promotion
Chunming He, Rihan Zhang, Fengyang Xiao, Dingming Zhang, Zhiwen Cao, Sina Farsiu
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 CurriSeg training to existing COD models | FEDER, FSEL, RUN (ResNet50, Res2Net50, PVT V2) · COD benchmarks | F-beta and composite gain: FEDER 0.715 -> 0.736 on COD10K F-beta; gains 2.13 to 3.94% | Improved |
| add CurriSeg training | RUN · COD training time | training time (h): 12.64 -> 8.32 | Improved |
| add CurriSeg training | RUN PVT V2 · CVC-ColonDB | mDice: 0.822 -> 0.845 | Improved |
| add CurriSeg training | RUN · GDD transparent objects | mIoU: 0.895 -> 0.902 | Improved |
| add warm-up curriculum alone, without uncertainty and temporal weighting (ablation) | FEDER · COD10K | F-beta: 0.715 -> 0.697 | Hurt |
| add pixel uncertainty weighting (PUE) on top of WCS (ablation) | CurriSeg · COD10K | F-beta: 0.697 -> 0.718 | Improved |
| add temporal statistics sample weighting (TSSW) (ablation) | CurriSeg · COD10K | F-beta: 0.718 -> 0.729 | Improved |
| add SBFT anti-curriculum phase after curriculum selection (ablation) | CurriSeg · COD10K | F-beta: 0.729 -> 0.736 | Improved |
| use SBFT alone without curriculum (ablation) | FEDER · COD10K | F-beta: 0.715 -> 0.723 | Improved |
| replace warm-up curriculum with self-paced learning or teacher-student curriculum (ablation) | CurriSeg · COD10K | F-beta: 0.738 and 0.735 vs 0.736 | No clear effect |
| use sqrt or squared warm-up schedule instead of linear p(t) (ablation) | CurriSeg · COD10K | F-beta: 0.730 and 0.726 vs 0.736 | Hurt |
| apply SBFT to all samples rather than only hard or random half (ablation) | CurriSeg · COD10K | F-beta: 0.730 / 0.728 -> 0.736 | Improved |
Closest papers by meaning
- Hunting Attributes: Context Prototype-Aware Learning for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Spider: A Unified Framework for Context-dependent Concept Segmentation (ICML 2024 · similarity 0.85)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.85)
- LENS: Learning to Segment Anything with Unified Reinforced Reasoning (AAAI 2026 · similarity 0.85)
- Distilling Spectral Graph for Object-Context Aware Open-Vocabulary Semantic Segmentation (CVPR 2025 · similarity 0.84)
- Perceptual Inductive Bias Is What You Need Before Contrastive Learning (CVPR 2025 · similarity 0.84)
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.
