Research map / Detection and segmentation / Open-set and few-shot detection
Dynamic-Structured Semantic Propagation Network
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 |
|---|---|---|---|
| Train one universal model jointly across datasets instead of dataset-specific fine-tuning | DSSPN (ResNet-101) · ADE20K val | mean IoU: 42.17 -> 43.68 | Helped |
| Train one universal model jointly across datasets instead of dataset-specific fine-tuning | DSSPN (ResNet-101) · COCO-Stuff test | mean IoU: 37.3 -> 38.9 | Helped |
| Train one universal model jointly across datasets instead of dataset-specific fine-tuning | DSSPN (ResNet-101) · Cityscapes test | IoU cla.: 74.6 -> 76.6 | Helped |
| Train one universal model jointly across datasets instead of dataset-specific fine-tuning | DSSPN (ResNet-101) · Mapillary val | mean IoU: 42.57 -> 45.01 | Helped |
| Use dynamic activation rather than a fixed neuron structure (ablation) | DSSPN · ADE20K val | mean IoU: 42.15 -> 42.03 | No clear effect |
| Add the dense semantic-enhanced block (ablation) | DSSPN · ADE20K val | mean IoU: 38.41 -> 42.03 | Helped |
| Concatenate ancestor features rather than summing them (ablation) | DSSPN dense block · ADE20K val | mean IoU: 41.74 -> 42.03 | Helped |
| Set dense-block hidden feature size to 32 rather than 48 (ablation) | DSSPN dense block · ADE20K val | mean IoU: 41.12 -> 42.03 | Hurt |
| Set dense-block hidden feature size to 64 rather than 48 (ablation) | DSSPN dense block · ADE20K val | mean IoU: 42.02 -> 42.03 | No clear effect |
| Use the original ADE20K synset hierarchy instead of the constructed hierarchy (ablation) | DSSPN · ADE20K val | mean IoU: 41.96 -> 42.03 | No clear effect |
| Use per-pixel binary cross-entropy rather than Softmax loss (ablation) | DSSPN · ADE20K val | mean IoU: 39.23 -> 42.03 | Helped |
Closest papers by meaning
- Semantic Projection Network for Zero- and Few-Label Semantic Segmentation (CVPR 2019 · similarity 0.88)
- Automated Label Unification for Multi-Dataset Semantic Segmentation with GNNs (NeurIPS 2024 · similarity 0.87)
- ReCo: Retrieve and Co-segment for Zero-shot Transfer (NeurIPS 2022 · similarity 0.87)
- Single-Stage Semantic Segmentation From Image Labels (CVPR 2020 · similarity 0.87)
- Flattening the Parent Bias: Hierarchical Semantic Segmentation in the Poincare Ball (CVPR 2024 · similarity 0.87)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 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.