Research map / Detection and segmentation / Open-set and few-shot detection
Exploit Visual Dependency Relations for Semantic Segmentation
Mingyuan Liu, Dan Schonfeld, Wei Tang
Check 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 intra-class, inter-class, and global dependency reasoning (ablation) | ResNet-50 segmentation network · Cityscapes validation | mIoU%: 71.38 -> 76.32 | Improved |
| Add intra-class dependency reasoning (ablation) | ResNet-50 segmentation network · Cityscapes validation | mIoU%: 71.38 -> 73.54 | Improved |
| Add inter-class dependency reasoning (ablation) | ResNet-50 segmentation network · Cityscapes validation | mIoU%: 71.38 -> 72.55 | Improved |
| Add global dependency reasoning (ablation) | ResNet-50 segmentation network · Cityscapes validation | mIoU%: 71.38 -> 74.92 | Improved |
| Add intermediate segmentation loss Lseg (ablation) | DependencyNet · Cityscapes validation | mIoU%: 76.32 -> 77.66 | Improved |
| Use group weighted convolutions instead of conventional convolutions for inter-class reasoning (ablation) | DependencyNet · Cityscapes validation | mIoU%: 76.14 -> 77.66 | Improved |
| Use Gconn rather than Gedge as the dependency graph (ablation) | DependencyNet inter-class reasoning · Cityscapes validation | mIoU%: 77.41 -> 77.66 | Improved |
| Use supervised category-specific representations rather than Class Center Mapping (ablation) | DependencyNet · Cityscapes validation | mIoU%: 74.61 -> 77.66 | Improved |
| Add dependency reasoning | ResNet-101 · BDD100K validation | mIoU%: 62.3 -> 63.9 | Improved |
| Add dependency reasoning | ResNet-101 · Cityscapes validation | mIoU%: 76.44 -> 79.15 | Improved |
| Add multi-scale inference (ablation) | DependencyNet with ResNet-101 · Cityscapes validation | mIoU%: 79.15 -> 80.48 | Improved |
| Add ASPP alongside DependencyNet and multi-scale inference (ablation) | DependencyNet with ResNet-101 · Cityscapes validation | mIoU%: 80.48 -> 82.01 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CCNet — CCNet: Criss-Cross Attention for Semantic Segmentation (ICCV 2019) · better on 1 of 1 shared result
- ACFNet — ACFNet: Attentional Class Feature Network for Semantic Segmentation (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
- Visual Dependency Transformers: Dependency Tree Emerges From Reversed Attention (CVPR 2023 · similarity 0.87)
- Context Prior for Scene Segmentation (CVPR 2020 · similarity 0.86)
- Deep Hierarchical Semantic Segmentation (CVPR 2022 · similarity 0.86)
- Visual Boundary Knowledge Translation for Foreground Segmentation (AAAI 2021 · similarity 0.86)
- IDRNet: Intervention-Driven Relation Network for Semantic Segmentation (NeurIPS 2023 · similarity 0.85)
- Regional Semantic Contrast and Aggregation for Weakly Supervised Semantic Segmentation (CVPR 2022 · 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.
