Research map / Detection and segmentation / CNN features
Improving Semantic Segmentation via Decoupled Body and Edge Supervision
Xiangtai Li, Xia Li, Li Zhang, Guangliang Cheng, Jianping Shi, Zhouchen Lin, Shaohua Tan, Yunhai Tong
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 decoupled body/edge framework | FCN ResNet-50 · Cityscapes val | mIoU (%): 76.6 -> 80.1 | Improved |
| add decoupled body/edge framework | FCN ResNet-101 · Cityscapes val | mIoU (%): 77.8 -> 80.8 | Improved |
| add L_body supervision (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 78.3 -> 78.8 | Improved |
| add L_bce alone (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 78.3 -> 78.3 | No clear effect |
| add edge OHEM loss with L_bce (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 78.3 -> 78.7 | Improved |
| add all three losses (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 78.3 -> 80.1 | Improved |
| add low-level F_fine to edge module (ablation) | ours · Cityscapes val | mIoU (%): 79.3 -> 80.1 | Improved |
| add hard pixel mining on edge loss (ablation) | ours · Cityscapes val | mIoU (%): 79.0 -> 80.1 | Improved |
| add warping in body generation (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 76.9 -> 78.3 | Improved |
| add encoder-decoder in body generation (ablation) | FCN + BG & EP · Cityscapes val | mIoU (%): 77.3 -> 78.3 | Improved |
| add edge preservation module (ablation) | FCN + BG · Cityscapes val | mIoU (%): 77.9 -> 78.3 | Improved |
| use our BG module instead of DCN | FCN · Cityscapes val | mIoU (%): 78.2 -> 80.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DANet — DANet: Divergent Activation for Weakly Supervised Object Localization (ICCV 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Drive&Segment: Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation (ECCV 2022 · similarity 0.89)
- Efficient Piecewise Training of Deep Structured Models for Semantic Segmentation (CVPR 2016 · similarity 0.89)
- Unsupervised Semantic Segmentation of Urban Scenes via Cross-Modal Distillation (IJCV 2025 · similarity 0.88)
- Multi-Scale Context Intertwining for Semantic Segmentation (ECCV 2018 · similarity 0.88)
- SUM Parts: Benchmarking Part-Level Semantic Segmentation of Urban Meshes (CVPR 2025 · similarity 0.87)
- Weakly Supervised Semantic Segmentation for Driving Scenes (AAAI 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.
