Research map / Detection and segmentation / CNN features
Devil Is in the Edges: Learning Semantic Boundaries From Noisy Annotations
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 |
|---|---|---|---|
| add the NMS loss and boundary thinning layer to CASENet (ablation) | CASENet · SBD re-annotated test set | MF (ODS): 63.20 -> 67.87 | Helped |
| add active alignment to the NMS-loss model (ablation) | STEAL · SBD re-annotated test set | MF (ODS): 67.87 -> 68.15 | Helped |
| add active alignment to the NMS-loss model (ablation) | STEAL · SBD original test set | MF (ODS): 65.93 -> 64.83 | Hurt |
| apply test-time NMS to predictions trained with the NMS loss (ablation) | STEAL · SBD original test set | MF (ODS): 64.15 -> 65.93 | Helped |
| apply test-time NMS to predictions trained with the NMS loss (ablation) | STEAL · SBD re-annotated test set | MF (ODS): 64.99 -> 67.87 | Helped |
| add active alignment during training (ablation) | STEAL · SBD high-quality test set, noisy train (+8px error) | MF (ODS): 50.58 -> 52.69; 29.32 -> 27.60 | Mixed |
| use STEAL coarse-to-fine refinement instead of the coarse input labels | STEAL · SBD simulated coarse labels, 16px error | IoU: 62.20 -> 82.89 | Helped |
| use STEAL coarse-to-fine refinement instead of the coarse input labels | STEAL · Cityscapes simulated coarse labels, 32px error | IoU: 19.44 -> 50.35 | Helped |
| train DeepLabV3+ with STEAL-refined train_extra annotations | DeepLabV3+ · Cityscapes val | mean IoU: 80.37 -> 80.55 | Helped |
Closest papers by meaning
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 2024 · similarity 0.89)
- Visual Boundary Knowledge Translation for Foreground Segmentation (AAAI 2021 · similarity 0.89)
- Beyond Pixel-wise Supervision: Local Structure Regularization for Semantic Segmentation (NeurIPS 2026 · similarity 0.88)
- Semantic Segmentation With Boundary Neural Fields (CVPR 2016 · similarity 0.88)
- Semantic Diffusion Network for Semantic Segmentation (NeurIPS 2022 · similarity 0.88)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.87)
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.