Research map / Label-efficient and robust learning / Semi-supervised learning
Keep Your Friends Close, and Your Enemies Farther: Distance-aware Voxel-wise Contrastive Learning for Semi-supervised Multi-organ Segmentation
Haochen Zhao, Jianwei Niu, Xuefeng Liu, Xiaozheng Xie, Li Kuang, Haotian Yang, Bin Dai, Hui Meng, Yong Wang
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 high-quality contrastive learning (HqCL) to CPS baseline (ablation) | CPS · FLARE 2022 (10% labeled) | mean Dice: 79.32 -> 82.00 | Improved |
| add DVCL loss on unreliable voxels on top of HqCL (ablation) | CPS + HqCL · FLARE 2022 (10% labeled) | mean Dice: 82.00 -> 83.41 | Improved |
| add neighbor partitioning strategy (NPS) (ablation) | CPS + HqCL + DVCL · FLARE 2022 (10% labeled) | mean Dice: 83.41 -> 84.09 | Improved |
| add query detection strategy (QDS) to exclude outlier queries (ablation) | CPS + HqCL + DVCL + NPS · FLARE 2022 (10% labeled) | mean Dice: 84.09 -> 84.15 | No clear effect |
| add all components (HqCL, DVCL, NPS, QDS) to CPS baseline (ablation) | CPS · FLARE 2022 (10% labeled) | mean Jaccard: 68.14 -> 73.62 | Improved |
| use DVCL log-likelihood loss instead of InfoNCE (ablation) | DVCL · FLARE 2022 (10% labeled) | mean Dice: 83.05 -> 83.41 | Improved |
| use DVCL log-likelihood loss instead of SimCLR (ablation) | DVCL · FLARE 2022 (10% labeled) | mean Dice: 82.77 -> 83.41 | Improved |
| use DVCL log-likelihood loss instead of NNCLR (ablation) | DVCL · FLARE 2022 (10% labeled) | mean Dice: 83.17 -> 83.41 | Improved |
| add SNCLR cosine-similarity weighting to log-likelihood loss (ablation) | DVCL log-likelihood · FLARE 2022 (10% labeled) | mean Dice: 83.41 -> 83.69 | No clear effect |
| add NPS weighting to log-likelihood loss instead of SNCLR weighting (ablation) | DVCL log-likelihood · FLARE 2022 (10% labeled) | mean Dice: 83.69 -> 84.09 | Improved |
| fill unreliable voxels via DVCL instead of discarding them (UGPCL) or complementary-label push (U2PL) | semi-supervised multi-organ segmentation · FLARE 2022 (10% labeled) | mean Dice: 82.1 -> 84.2 | Improved |
| DVCL vs second-best method at 10% labels on AMOS | DVCL · AMOS (10% labeled) | mean Dice: 65.8 -> 71.4 | Improved |
Closest papers by meaning
- Rethinking Semi-Supervised Medical Image Segmentation: A Variance-Reduction Perspective (NeurIPS 2023 · similarity 0.89)
- Neighbor Does Matter: Density-Aware Contrastive Learning for Medical Semi-supervised Segmentation (AAAI 2025 · similarity 0.88)
- A Dual-space Patch-driven Complementary Learning Framework for Semi-supervised Multi-organ Segmentation (ECCV 2026 · similarity 0.88)
- DyCON: Dynamic Uncertainty-aware Consistency and Contrastive Learning for Semi-supervised Medical Image Segmentation (CVPR 2025 · similarity 0.86)
- A Semantic Knowledge Complementarity based Decoupling Framework for Semi-supervised Class-imbalanced Medical Image Segmentation (CVPR 2025 · similarity 0.85)
- DuSSS: Dual Semantic Similarity-Supervised Vision-Language Model for Semi-Supervised Medical Image Segmentation (AAAI 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.
