Research map / Detection and segmentation / CNN features
E-CRF: Embedded Conditional Random Field for Boundary-caused Class Weights Confusion in 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 |
|---|---|---|---|
| fuse CRF into CNN end-to-end as E-CRF (vs. Vanilla-CRF offline post-processing) | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): +1.42 vs. +0.54 for Vanilla-CRF | Helped |
| fuse CRF into CNN end-to-end as E-CRF (vs. Joint-CRF) | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): +1.42 vs +1.01 for Joint-CRF | Helped |
| add pairwise message passing module psi_p (E-CRF pairwise only) (ablation) | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): +0.87 | Helped |
| add superpixel module psi_s (ablation) | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): +1.25 | Helped |
| add both psi_p and psi_s (full E-CRF) (ablation) | DeeplabV3+ ResNet-50 · ADE20K val | mIoU (%): +1.48 over baseline; +0.29 over pairwise-only, +0.37 over superpixel-only | Helped |
| add Non-local attention (ablation) | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): +0.74 (less than E-CRF's +1.42) | Helped |
| fuse CRF into SegFormer (E-CRF) | SegFormer MiT-B5 · ADE20K val | mIoU (%): +1.01 | Helped |
| fuse CRF into SegFormer (E-CRF) | SegFormer MiT-B5 · Cityscapes val | mIoU (%): +0.82 | Helped |
| offline Vanilla-CRF post-processing | DeeplabV3+ ResNet-101 · Cityscapes val | mIoU (%): +0.07 | No clear effect |
| superpixel count set to 200 vs. 100/300/400 (with psi_p) (ablation) | E-CRF ResNet-50 · ADE20K val | mIoU (%): 44.20 at 200 vs 44.02, 44.13, 43.96 at other counts | Mixed |
| E-CRF vs SegFix boundary refinement | DeeplabV3+ ResNet-101 · ADE20K val | mIoU (%): 46.02 vs 45.62 for SegFix | Helped |
Closest papers by meaning
- Efficient Linear Programming for Dense CRFs (CVPR 2017 · similarity 0.90)
- Semantic Image Segmentation With Task-Specific Edge Detection Using CNNs and a Discriminatively Trained Domain Transform (CVPR 2016 · similarity 0.89)
- Efficient Piecewise Training of Deep Structured Models for Semantic Segmentation (CVPR 2016 · similarity 0.88)
- Semantic Segmentation via Structured Patch Prediction, Context CRF and Guidance CRF (CVPR 2017 · similarity 0.88)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- W2P: Switching from Weak Supervision to Partial Supervision for Semantic 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.