Research map / Detection and segmentation / CNN features
SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers
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 |
|---|---|---|---|
| Scale the encoder from MiT-B0 to MiT-B5 (ablation) | SegFormer · ADE20K | mIoU: 37.4 -> 51.0 | Helped |
| Increase decoder dimension from C=256 to C=768 (ablation) | SegFormer All-MLP decoder · ADE20K | mIoU: 44.9 -> 45.4 | Helped |
| Increase decoder dimension beyond C=768 to C=1024 (ablation) | SegFormer All-MLP decoder · ADE20K | mIoU: 45.4 -> 45.2 | Hurt |
| Replace positional encoding with Mix-FFN (ablation) | SegFormer encoder · Cityscapes at 768×768 inference | mIoU: 77.3 -> 80.5 | Helped |
| Replace positional encoding with Mix-FFN (ablation) | SegFormer encoder · Cityscapes at 1024×2048 inference | mIoU: 74.0 -> 79.8 | Helped |
| Change test resolution from 768×768 to 1024×2048 (ablation) | Positional encoding encoder · Cityscapes | mIoU: 77.3 -> 74.0 | Hurt |
| Change test resolution from 768×768 to 1024×2048 (ablation) | Mix-FFN encoder · Cityscapes | mIoU: 80.5 -> 79.8 | Hurt |
| Fuse all encoder stages instead of using only stage 4 (ablation) | MiT-B2 with All-MLP decoder · ADE20K | mIoU: 43.1 -> 45.4 | Helped |
| Replace the Transformer encoder with a CNN encoder (ablation) | All-MLP decoder on ADE20K · ADE20K | mIoU: 45.4 -> 34.7 | Hurt |
| Use Mapillary Vistas pre-training | SegFormer-B5 · Cityscapes test | mIoU: 82.2 -> 83.1 | Helped |
Closest papers by meaning
- FeedFormer: Revisiting Transformer Decoder for Efficient Semantic Segmentation (AAAI 2023 · similarity 0.91)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.91)
- SED: A Simple Encoder-Decoder for Open-Vocabulary Semantic Segmentation (CVPR 2024 · similarity 0.88)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.87)
- Learning Content-Enhanced Mask Transformer for Domain Generalized Urban-Scene Segmentation (AAAI 2024 · similarity 0.86)
- SegMAN: Omni-scale Context Modeling with State Space Models and Local Attention for Semantic Segmentation (CVPR 2025 · 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.