Research map / Detection and segmentation / CNN features
Revisiting Efficient Semantic Segmentation: Learning Offsets for Better Spatial and Class Feature Alignment
Shi-Chen Zhang, Yunheng Li, Yu-Huan Wu, Qibin Hou, Ming-Ming Cheng
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 offset learning | SegNeXt-T · ADE20K | mIoU: 41.1 -> 43.0 | Improved |
| add offset learning | SegNeXt-T · Cityscapes | mIoU: 79.8 -> 80.0 | Improved |
| add offset learning | SegNeXt-B · COCO-Stuff | mIoU: 45.8 -> 45.8 | No clear effect |
| add offset learning | SegNeXt-B · ADE20K | mIoU: 48.5 -> 49.4 | Improved |
| add offset learning | SegFormer-B0 · ADE20K | mIoU: 37.4 -> 40.1 | Improved |
| add offset learning | SegFormer-B5 · ADE20K | mIoU: 49.1 -> 50.6 | Improved |
| add offset learning | SegFormer-B5 · COCO-Stuff | mIoU: 46.7 -> 47.2 | Improved |
| add offset learning | Mask2Former-Tiny · ADE20K | mIoU: 47.7 -> 50.3 | Improved |
| add FreqFusion (ablation) | OffSeg-T · ADE20K | mIoU: 40.7 -> 42.0 | Improved |
| add class offset learning (ablation) | OffSeg-T with FreqFusion · ADE20K | mIoU: 42.0 -> 42.9 | Improved |
| add feature offset learning (ablation) | OffSeg-T with FreqFusion · ADE20K | mIoU: 42.0 -> 43.5 | Improved |
| add both offset branches (ablation) | OffSeg-T with FreqFusion · ADE20K | mIoU: 42.0 -> 44.2 | Improved |
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.
- RTFormer-Base — RTFormer: Efficient Design for Real-Time Semantic Segmentation with Transformer (NeurIPS 2022) · 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
- Endpoints Weight Fusion for Class Incremental Semantic Segmentation (CVPR 2023 · similarity 0.88)
- ReBac: Current-to-Past Residual Background Correction for Class-Incremental Semantic Segmentation (IJCV 2026 · similarity 0.87)
- Region-Based Semantic Segmentation with End-to-End Training (ECCV 2016 · similarity 0.86)
- Towards Stable Source-Free Domain Adaptive Semantic Segmentation (IJCV 2026 · similarity 0.85)
- Rethinking Alignment and Uniformity in Unsupervised Image Semantic Segmentation (AAAI 2023 · similarity 0.85)
- Improving Semi-Supervised Semantic Segmentation with Sliced-Wasserstein Feature Alignment and Uniformity (CVPR 2025 · similarity 0.84)
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.
