Research map / Detection and segmentation / CNN features
Learning Dynamic Routing for Semantic Segmentation
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 |
|---|---|---|---|
| Use input-dependent dynamic routing instead of a fixed architecture (ablation) | Segmentation network · Cityscapes val | mIoU(%): 74.5 -> 76.1 | Helped |
| Use dynamic routing rather than a static common connection pattern | Dynamic-A versus Common-A · Cityscapes val | mIoU(%): 71.6 -> 72.8 | Helped |
| Use dynamic routing rather than a static common connection pattern | Dynamic-B versus Common-B · Cityscapes val | mIoU(%): 73.0 -> 73.8 | Helped |
| Use dynamic routing rather than a static common connection pattern | Dynamic-C versus Common-C · Cityscapes val | mIoU(%): 73.2 -> 74.6 | Helped |
| Use two SepConv3×3 operations per cell rather than one (ablation) | Dynamic routing · Cityscapes val | mIoU(%): 71.2 -> 76.1 | Helped |
| Increase cell depth from two to three SepConv3×3 operations (ablation) | Dynamic routing · Cityscapes val | mIoU(%): 76.1 -> 75.2 | Hurt |
| Replace the proposed max(0, Tanh) gate activation with Softmax (ablation) | Dynamic routing · Cityscapes val | mIoU(%): 76.1 -> 74.1 | Hurt |
| Replace the proposed max(0, Tanh) gate activation with Sigmoid (ablation) | Dynamic routing · Cityscapes val | mIoU(%): 76.1 -> 75.9 | Hurt |
| Apply a resource budget rather than leave routing unconstrained (ablation) | Dynamic routing · Cityscapes val | mIoU(%): 76.1 -> 74.6 | Hurt |
| Apply a resource budget rather than leave routing unconstrained (ablation) | Dynamic routing · Cityscapes val | FLOPs(G): 119.5 -> 66.6 | Helped |
| Increase training from 186K to 372K iterations (ablation) | Dynamic network · Cityscapes val, single scale, no flipping | mIoU(%): 76.1 -> 77.4 | Helped |
| Add Scheduled Drop Path (ablation) | Dynamic network · Cityscapes val, single scale, no flipping | mIoU(%): 77.4 -> 78.3 | Helped |
Closest papers by meaning
- Searching Efficient Semantic Segmentation Architectures via Dynamic Path Selection (NeurIPS 2025 · similarity 0.89)
- Coupled Segmentation and Edge Learning via Dynamic Graph Propagation (NeurIPS 2021 · similarity 0.88)
- Graph-Guided Architecture Search for Real-Time Semantic Segmentation (CVPR 2020 · similarity 0.87)
- Semantic Video Segmentation by Gated Recurrent Flow Propagation (CVPR 2018 · similarity 0.87)
- Hierarchically Gated Deep Networks for Semantic Segmentation (CVPR 2016 · similarity 0.86)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time 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.