Research map / Detection and segmentation / CNN features
Specialize and Fuse: Pyramidal Output Representation for Semantic Segmentation
Chi-Wei Hsiao, Cheng Sun, Hwann-Tzong Chen, Min Sun
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 pyramidal output instead of single output (no contextual module) (ablation) | segmentation head · ADE20K subset (HRNet32) | mIoU: 40.42 -> 42.07 | Improved |
| add pyramidal output instead of single output (with ANL) (ablation) | segmentation head · ADE20K subset (HRNet32) | mIoU: 42.02 -> 43.07 | Improved |
| add pyramidal output instead of single output (with proposed contextual module) (ablation) | segmentation head · ADE20K subset (HRNet32) | mIoU: 42.00 -> 44.31 | Improved |
| add coarse-to-fine contextual module to single output (ablation) | single-output baseline · ADE20K subset (HRNet32) | mIoU: 40.42 -> 42.00 | Improved |
| add coarse-to-fine contextual module to pyramidal output (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 42.07 -> 44.31 | Improved |
| add single ANL to pyramidal output (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 42.07 -> 43.07 | Improved |
| add ANL at all pyramid levels (ANL-multi) (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 42.07 -> 43.45 | Improved |
| train all levels on full map without relabeling (naive) (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 40.42 -> 40.41 | No clear effect |
| relabel descendants of ground-truth unity-cells as don't care (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 40.41 -> 41.54 | Improved |
| relabel descendants of true-positive unity-cells instead of ground truth (ablation) | pyramidal output · ADE20K subset (HRNet32) | mIoU: 41.54 -> 42.07 | Improved |
| use multi-level supervision as auxiliary only (no fuse) (ablation) | single output + contextual module · ADE20K subset (HRNet32) | mIoU: 42.00 -> 42.63 | Improved |
| use specialize and fuse rather than auxiliary supervision (ablation) | single output + contextual module · ADE20K subset (HRNet32) | mIoU: 42.63 -> 44.31 | Improved |
Closest papers by meaning
- Laplacian Pyramid Reconstruction and Refinement for Semantic Segmentation (ECCV 2016 · similarity 0.87)
- SED++: A Simple Encoder-Decoder for Improved Open-Vocabulary Semantic Segmentation (TPAMI 2025 · similarity 0.86)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.86)
- Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation (ICML 2025 · similarity 0.86)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.85)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · 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.
