Research map / Detection and segmentation / CNN features
RANet: Region Attention Network 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 RCB and RIB region attention to the backbone (ablation) | ResNet-101 backbone · Cityscapes val | mIoU: 77.5 -> 81.9 | Helped |
| Use region-wise context instead of SPP pixel-wise context (ablation) | RANet · Cityscapes val | mIoU: 79.6 -> 81.9 | Helped |
| Use region-wise context instead of non-local attention (ablation) | RANet · Cityscapes val | mIoU: 80.5 -> 81.9 | Helped |
| Combine boundary and semantic score maps for region attention instead of boundary map only (ablation) | RCB · Cityscapes val | mIoU: 80.8 -> 81.9 | Helped |
| Combine boundary and semantic score maps instead of semantic map only (ablation) | RCB · Cityscapes val | mIoU: 81.3 -> 81.9 | Helped |
| Add intra-region collection and distribution (ablation) | RIB · Cityscapes val | mIoU: 79.2 -> 81.9 | Helped |
| Add inter-region interaction (ablation) | RIB · Cityscapes val | mIoU: 81.1 -> 81.9 | Helped |
| Increase representative pixels per region from 2 to 8 (ablation) | RIB · Cityscapes val | mIoU: 81.2 -> 81.9 | Helped |
| Increase representative pixels per region from 8 to 12 (ablation) | RIB · Cityscapes val | mIoU: 81.9 -> 81.6 | Hurt |
| Use full RANet instead of coarse semantic segmentation (ablation) | RANet · Cityscapes val / PASCAL Context / COCO-Stuff | mIoU: 77.3 -> 81.9; 51.2 -> 54.9; 37.9 -> 40.7 | Helped |
| Use RANet region attention instead of Asymmetric NL representative-pixel selection | Attention module · Cityscapes val | mIoU: 81.0 -> 81.9 | Helped |
| Add extra coarse annotations to training | RANet (HRNetV2-W48+ASPP) · Cityscapes test | mIoU: 83.4 -> 84.0 | Helped |
Closest papers by meaning
- Squeeze-and-Attention Networks for Semantic Segmentation (CVPR 2020 · similarity 0.88)
- AttaNet: Attention-Augmented Network for Fast and Accurate Scene Parsing (AAAI 2021 · similarity 0.88)
- Fully Attentional Network for Semantic Segmentation (AAAI 2022 · similarity 0.87)
- Attention to Scale: Scale-Aware Semantic Image Segmentation (CVPR 2016 · similarity 0.87)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.85)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · 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.