Research map / Detection and segmentation / CNN features
Not All Pixels are Equal: Learning Pixel Hardness for Semantic Segmentation
Xin Xiao, Daiguo Zhou, Jiagao Hu, Yi Hu, Yongchao Xu
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 hardness level learning (HL) | UPerNet ResNet-18 · Cityscapes val | mIoU: 74.39 -> 76.10 | Improved |
| add HL | BiSeNetV2 · Cityscapes val | mIoU: 72.92 -> 75.27 | Improved |
| add HL | UPerNet ConvNeXt-Base · Cityscapes val | mIoU: 80.89 -> 81.33 | Improved |
| add HL | PSPNet ResNet-18 · iSAID val | mIoU: 60.76 -> 62.93 | Improved |
| add HL | DeepLabV3+ ResNet-101 · ADE20K val | mIoU: 45.12 -> 45.43 | Improved |
| add HL | FCN HRNet-W18 (small) · Total-Text test | fgIoU: 66.62 -> 68.56 | Improved |
| add HL to U2PL | U2PL (w/CutMix) · Cityscapes semi-supervised 1/16 | mIoU: 71.11 -> 72.62 | Improved |
| add HL | DeepLabV3+ ResNet-101 · GTAV to Cityscapes | mIoU: 36.56 -> 43.06 | Improved |
| use HL pixel weights instead of cross-entropy (ablation) | PSPNet ResNet-18 · Cityscapes val | mIoU: 74.30 -> 75.60 | Improved |
| use focal loss pixel weights instead of cross-entropy (ablation) | PSPNet ResNet-18 · Cityscapes val | mIoU: 74.30 -> 73.35 | Hurt |
| use balanced CE instead of cross-entropy (ablation) | PSPNet ResNet-101 · Cityscapes val | mIoU: 79.60 -> 79.16 | Hurt |
| use OHEM CE instead of cross-entropy (ablation) | PSPNet ResNet-101 · Cityscapes val | mIoU: 79.60 -> 79.64 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BiSeNetV2 — BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (IJCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Exploring Cross-Image Pixel Contrast for Semantic Segmentation (ICCV 2021 · similarity 0.89)
- Rethinking Self-Supervised Semantic Segmentation: Achieving End-to-End Segmentation (TPAMI 2024 · similarity 0.88)
- PSDPM: Prototype-based Secondary Discriminative Pixels Mining for Weakly Supervised Semantic Segmentation (CVPR 2024 · similarity 0.87)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.86)
- Diffusion-Driven Two-Stage Active Learning for Low-Budget Semantic Segmentation (NeurIPS 2025 · similarity 0.86)
- When Semantic Segmentation Meets Frequency Aliasing (ICLR 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.
