Research map / Detection and segmentation / CNN features
Transformer Scale Gate for Semantic Segmentation
Hengcan Shi, Munawar Hayat, Jianfei Cai
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 Transformer Scale Gate | Swin-Tiny segmentation model · Pascal Context validation | mIoU (%): 50.2 -> 54.5 (+4.3) | Improved |
| Add Transformer Scale Gate | Swin-Large segmentation model · ADE20K validation | mIoU (%): 52.1 -> 54.2 (+2.1) | Improved |
| Add encoder TSGE to replace conventional feature refinement (ablation) | Swin-Tiny with UperNet · Pascal Context validation | mIoU (%): 50.2 -> 51.6 (+1.4) | Improved |
| Use self-attention gates to weight multi-scale features (ablation) | Swin-Tiny with baseline decoder · Pascal Context validation | mIoU (%): 51.3 -> 52.4 (+1.1) | Improved |
| Add TSGD to the baseline decoder (ablation) | Swin-Tiny · Pascal Context validation | mIoU (%): 51.3 -> 52.7 (+1.4) | Improved |
| Replace FPN feature refinement with TSGE (ablation) | Swin-Tiny with baseline decoder · Pascal Context validation | mIoU (%): 52.9 -> 53.9 (+1.0) | Improved |
| Add TSGD to the FPN-refined encoder (ablation) | Swin-Tiny + FPN · Pascal Context validation | mIoU (%): 52.9 -> 54.1 (+1.2) | Improved |
| Combine TSGE and TSGD (ablation) | Swin-Tiny segmentation model · Pascal Context validation | mIoU (%): 50.2 -> 54.5 (+4.3) | Improved |
| Selectively fuse all scales rather than combine all four scales without selection (ablation) | Swin-Tiny with encoder and decoder · Pascal Context validation | mIoU (%): 52.9 -> 54.5 | Improved |
| Concatenate multi-head attention maps instead of averaging them (ablation) | TSG · Pascal Context validation | mIoU (%): 54.1 -> 54.5 (+0.4) | Improved |
| Use independent rather than shared weights across TSG modules (ablation) | TSG modules across encoder steps and decoder blocks · Pascal Context validation | mIoU (%): 53.8 -> 54.5 (+0.7) | Improved |
Closest papers by meaning
- Segmenter: Transformer for Semantic Segmentation (ICCV 2021 · similarity 0.88)
- Transformer-Based Visual Segmentation: A Survey (TPAMI 2024 · similarity 0.88)
- FeedFormer: Revisiting Transformer Decoder for Efficient Semantic Segmentation (AAAI 2023 · similarity 0.88)
- A Transformer-Based Decoder for Semantic Segmentation with Multi-level Context Mining (ECCV 2022 · similarity 0.88)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.87)
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.87)
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.
