Research map / Detection and segmentation / CNN features
A Transformer-Based Decoder for Semantic Segmentation with Multi-level Context Mining
Bowen Shi, Dongsheng Jiang, Xiaopeng Zhang, Han Li, Wenrui Dai, Junni Zou, Hongkai Xiong, Qi Tian
This paper tried a common change: Add attention module
add the second attention layer Y_exter in external mining · ADE20K val · mIoU: 0.51% performance loss when removed Improved
What 492 accepted papers report for this change
- 444 improved
- 20 no clear effect
- 4 mixed
- 24 hurt
- Multi-Scale Wavelet Transformers for Operator Learning of Dynamical Systems ICML 2026 · add wavelet attention block · CKF step 64 · Rel L2: 1.09 -> 0.35Improved
- DA-DFGAS:Differentiable Federated Graph Neural Architecture Search with Distribution-Aware Attentive Aggregation AAAI 2026 · add attention mechanism to aggregation · Table 4 datasets · parameters: about 5-10% more parametersHurt
- Caracal: Causal Architecture via Spectral Mixing ICML 2026 · add sliding window attention layers to MHF stack · Avg accuracy · avg accuracy (%): 48.22 -> 49.01Improved
See this change in detailMatched by the wording of each paper's own result, read by AI from the full text. Papers report gains more often than failures, so this shows what gets published, not the chance a change works.
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 one-head internal self-attention in the decoder (ablation) | SegFormer-B1 · ADE20K val | mIoU: 40.97 -> 42.65 | Improved |
| use more heads, deeper or MLP in internal attention (ablation) | SegFormer-B1 + internal attention · ADE20K val | mIoU: 42.65 -> 42.53 | No clear effect |
| add MLP after internal attention (ablation) | SegFormer-B1 + internal attention · ADE20K val | mIoU: 42.65 -> 42.81 | Improved |
| use momentum-updated external tokens instead of none (ablation) | SegFormer-B1 · ADE20K val | mIoU: 40.97 -> 41.44 | Improved |
| use learnable instead of momentum external tokens with mid-level supervision (ablation) | SegFormer-B1 · ADE20K val | mIoU: 41.74 -> 42.37 | Improved |
| add category-specific loss L_attn to learnable external tokens (ablation) | SegFormer-B1 · ADE20K val | mIoU: 41.57 -> 42.37 | Improved |
| add decoupling of external tokens (ablation) | SegFormer-B1 · ADE20K val | mIoU: 42.37 -> 42.84 | Improved |
| add the second attention layer Y_exter in external mining (ablation) | SegFormer-B1 · ADE20K val | mIoU: 0.51% performance loss when removed | Improved |
| add internal and external mining together (ablation) | SegFormer-B1 · ADE20K val | mIoU: 40.97 -> 44.05 | Improved |
| add SegDeformer to SegFormer-B2 | SegFormer-B2 · ADE20K val | mIoU: 45.58 -> 47.49 | Improved |
| deepen the base decoder instead of using SegDeformer (ablation) | SegFormer-B2 · ADE20K val | mIoU: 45.58 -> 45.68 | No clear effect |
| replace UperNet with SegDeformer on Deit-S | Deit-S · ADE20K val | mIoU: 42.87 -> 43.93 | Improved |
Closest papers by meaning
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.93)
- Segmenter: Transformer for Semantic Segmentation (ICCV 2021 · similarity 0.92)
- 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 0.89)
- A Simple Image Segmentation Framework via In-Context Examples (NeurIPS 2024 · similarity 0.87)
- SegMAN: Omni-scale Context Modeling with State Space Models and Local Attention for Semantic Segmentation (CVPR 2025 · 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.
