Research map / Detection and segmentation / CNN features
Active Boundary Loss for Semantic Segmentation
Read the paperCheck 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 IABL to OCR | OCR (HRNetV2-W48) · ADE20K validation | mIoU: 45.66 -> 46.88 | Helped |
| Add IABL to UperNet | UperNet (Swin-B) · ADE20K validation | mIoU: 51.66 -> 52.40 | Helped |
| Add IABL to OCR | OCR (HRNetV2-W48) · Cityscapes validation | mIoU: 82.2 -> 82.9 | Helped |
| Add ABL to CE + IoU training (ablation) | DeepLabV3 · Cityscapes validation | mIoU: 80.2 -> 80.5 | Helped |
| Use ABL only in the last 20% of epochs, without IoU loss (ablation) | DeepLabV3 · Cityscapes validation | mIoU: 79.5 -> 79.6 | No clear effect |
| Remove gradient detachment from IABL (ablation) | DeepLabV3 · Cityscapes validation | mIoU: 80.5 -> 76.0 | Hurt |
| Replace active boundary loss with full KL-divergence loss (ablation) | OCR · ADE20K | mIoU: 45.38 -> 45.11 | Hurt |
| Use a 2% predicted-boundary-pixel threshold instead of 1% (ablation) | FCN (HRNetV2-W18s) · Cityscapes validation | mIoU: 75.59% -> 75.46% | Hurt |
| Use a 0.5% predicted-boundary-pixel threshold instead of 1% (ablation) | FCN (HRNetV2-W18s) · Cityscapes validation | mIoU: 75.59% -> 75.41% | Hurt |
| Decrease IoU-loss weight while increasing ABL weight during training (ablation) | FCN (HRNetV2-W18s) · Cityscapes validation | mIoU: 75.59% -> 75.65% | No clear effect |
| Add IABL to STM fine-tuning | STM · DAVIS-2016 validation | J-mean: 89.08 -> 89.29 | Helped |
| Add IABL to STM fine-tuning | STM · DAVIS-2016 validation | F-mean: 90.66 -> 90.82 | Helped |
Closest papers by meaning
- InverseForm: A Loss Function for Structured Boundary-Aware Segmentation (CVPR 2021 · similarity 0.91)
- Devil Is in the Edges: Learning Semantic Boundaries From Noisy Annotations (CVPR 2019 · similarity 0.88)
- Joint Semantic Segmentation and Boundary Detection Using Iterative Pyramid Contexts (CVPR 2020 · similarity 0.87)
- Improving Semantic Segmentation via Video Propagation and Label Relaxation (CVPR 2019 · similarity 0.87)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic Segmentation (AAAI 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.