Research map / Detection and segmentation / Weakly supervised detection and segmentation
Noisy Boundaries: Lemon or Lemonade for Semi-Supervised Instance Segmentation?
Check 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 the full semi-supervised method with pseudo labels, NTM, and BPM | Mask R-CNN instance segmentation · Cityscapes, 30% labeled | mask AP: 28.3 -> 32.4 | Helped |
| Use the full semi-supervised method with pseudo labels, NTM, and BPM | Mask R-CNN instance segmentation · COCO, 5% labeled | mask AP: 17.3 -> 24.9 | Helped |
| Use the full semi-supervised method with pseudo labels, NTM, and BPM | Mask R-CNN instance segmentation · BDD100K, 7k masks and 67k unlabeled images | AP: 21.6 -> 26.3 | Helped |
| Add data augmentation during student training (ablation) | Semi-supervised instance segmentation · Cityscapes, 30% labeled + 70% unlabeled | mask AP: 28.3 -> 30.2 | Helped |
| Add data augmentation during student training (ablation) | Fully supervised instance segmentation · Cityscapes, 30% labeled | mask AP: 26.3 -> 26.6 | No clear effect |
| Add the noise-tolerant mask head (ablation) | Semi-supervised instance segmentation with data augmentation · Cityscapes, 30% labeled + 70% unlabeled | mask AP: 30.2 -> 31.1 | Helped |
| Add the noise-tolerant mask head (ablation) | Semi-supervised instance segmentation with data augmentation · Cityscapes, 30% labeled + 70% unlabeled | boundary AP: 10.4 -> 10.9 | Helped |
| Add the boundary-preserving map (ablation) | Semi-supervised instance segmentation with data augmentation · Cityscapes, 30% labeled + 70% unlabeled | mask AP: 30.2 -> 31.0 | Helped |
| Add the boundary-preserving map (ablation) | Semi-supervised instance segmentation with data augmentation · Cityscapes, 30% labeled + 70% unlabeled | boundary AP: 10.4 -> 11.6 | Helped |
| Use focal loss for the detection branch | Proposed semi-supervised instance segmentation · Cityscapes, 40% labeled | mask AP: 33.0 -> 34.1 | Helped |
| Use BPM together with NTM rather than either alone (ablation) | Semi-supervised instance segmentation with data augmentation · Cityscapes, 30% labeled + 70% unlabeled | boundary AP: 11.6 -> 11.6 | No clear effect |
Closest papers by meaning
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.90)
- Robust Pseudo-Labeling via Decoupled Class-Aware Filtering and Dynamic Category Correction (AAAI 2026 · similarity 0.90)
- P2Seg: Pointly-supervised Segmentation via Mutual Distillation (ICLR 2024 · similarity 0.88)
- WISH: Weakly Supervised Instance Segmentation using Heterogeneous Labels (CVPR 2025 · similarity 0.88)
- Extreme Point Supervised Instance Segmentation (CVPR 2024 · similarity 0.87)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.86)
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.