Research map / Detection and segmentation / CNN features
RTFormer: Efficient Design for Real-Time Semantic Segmentation with Transformer
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 |
|---|---|---|---|
| replace multi-head external attention (EA+EA r=1) with GPU-Friendly Attention (ablation) | RTFormer-Slim · ADE20K val | mIoU: 32.2 -> 32.8 | Helped |
| replace multi-head self attention with GPU-Friendly Attention (ablation) | RTFormer-Slim · ADE20K val | FPS: 97.4 -> 189.8 | Helped |
| replace multi-head self attention with GPU-Friendly Attention (ablation) | RTFormer-Slim · ADE20K val | mIoU: 32.7 -> 32.8 | No clear effect |
| add cross-resolution attention in high-res branch (GFA+GFA -> GFA+CA) (ablation) | RTFormer-Slim · ADE20K val | mIoU: 32.8 -> 33.0 | Helped |
| replace typical FFN with two 3x3 conv FFN (ablation) | RTFormer-Slim · ADE20K val | mIoU: 32.15 -> 33.0 | Helped |
| use grouped double normalization (8,2 groups) instead of original double normalization (ablation) | GFA+GFA · ADE20K val | mIoU: 32.2 -> 32.8 | Helped |
| increase cross-feature size from 8x8 to 12x12 (ablation) | RTFormer-Slim · ADE20K val | mIoU: 33.00 -> 32.94 | No clear effect |
| replace DDRNet-23 with RTFormer-Base | DDRNet-23 · CamVid test | mIoU: 76.3 -> 82.5 | Helped |
| replace DDRNet-23 with RTFormer-Base | DDRNet-23 · COCOStuff test | mIoU: 32.1 -> 35.3 | Helped |
Closest papers by meaning
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.89)
- Rethinking Semantic Segmentation From a Sequence-to-Sequence Perspective With Transformers (CVPR 2021 · similarity 0.89)
- SCTNet: Single-Branch CNN with Transformer Semantic Information for Real-Time Segmentation (AAAI 2024 · similarity 0.88)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.87)
- DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation (ICLR 2024 · similarity 0.85)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · 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.