Research map / Detection and segmentation / CNN features
SegFormer: A Topic Segmentation Model with Controllable Range of Attention
Haitao Bai, Pinghui Wang, Ruofei Zhang, Zhou Su
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 |
|---|---|---|---|
| Use SegFormer instead of Transformer2 | WIKI-SECTION · En Disease, De Disease, En City, and De City | Pk: 18.8 -> 17.6; 16.0 -> 14.9; 9.1 -> 8.2; 7.3 -> 6.8 | Improved |
| Add topic classification loss to guide context range (ablation) | SegFormer · WIKI-SECTION | Pk: 22.7 -> 17.6; 18.7 -> 14.9; 8.6 -> 8.2; 7.1 -> 6.8 | Improved |
| Add the sentence contextualization encoder (ablation) | SegFormer · WIKI-SECTION | Pk: 23.3 -> 17.6; 18.6 -> 14.9; 10.6 -> 8.2; 11.1 -> 6.8 | Improved |
| Add training and iterative inference strategy (ablation) | SegFormer · WIKI-SECTION | Pk: 18.4 -> 17.6; 15.6 -> 14.9; 8.6 -> 8.2; 7.2 -> 6.8 | Improved |
| Add context aggregator along with training and inference strategy (ablation) | SegFormer · WIKI-SECTION | Pk: 23.7 -> 17.6; 19.9 -> 14.9; 9.0 -> 8.2; 8.2 -> 6.8 | Improved |
| Add topic loss, context aggregator, and training/inference strategy (ablation) | SegFormer · WIKI-SECTION | Pk: 26.6 -> 17.6; 22.9 -> 14.9; 9.7 -> 8.2; 8.5 -> 6.8 | Improved |
| Use SegFormer rather than the strongest listed baseline | WIKI-SECTION transfer datasets · WIKI-50, Cities, Elements, Clinical Books | Pk: 29.3 -> 25.3; 16.1 -> 15.2; 45.2 -> 49.4; 36.9 -> 28.6 | Mixed |
Later papers that compared with this method
33 later accepted papers report SegFormer as a baseline in their results tables; 31 of them report a better number on most of the results they share with it.
- Frequency Mining Empowered by Text Aggregation: A New Perspective on Document Image Tampering Detection (AAAI 2026 · better on 4 of 4 shared results)
- MambaSeg: Harnessing Mamba for Accurate and Efficient Image-Event Semantic Segmentation (AAAI 2026 · better on 4 of 4 shared results)
- Restoration Adaptation for Semantic Segmentation on Low Quality Images (IJCV 2026 · better on 4 of 4 shared results)
- RMAE-ProGRess: Advancing Semantic Segmentation in Unstructured Environments (CVPR 2026 · better on 3 of 3 shared results)
- Visual-Geometric Collaborative Guidance for Affordance Learning (IJCV 2026 · better on 3 of 3 shared results)
- E²I-VRWKV: Explicit EPI-Representation and Interaction-Aware Vision-RWKV for Light Field Semantic Segmentation (ICML 2026 · better on 2 of 2 shared results)
- Urban Socio-Semantic Segmentation with Vision-Language Reasoning (ICLR 2026 · better on 2 of 2 shared results)
- Hilbert Curve-Based Attention Enabling Topology-Preserving Image Tensor Representation for Semantic Segmentation Network (CVPR 2026 · better on 2 of 2 shared results)
25 more not listed.
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
- textTOvec: DEEP CONTEXTUALIZED NEURAL AUTOREGRESSIVE TOPIC MODELS OF LANGUAGE WITH DISTRIBUTED COMPOSITIONAL PRIOR (ICLR 2019 · similarity 0.84)
- Neural Dynamic Focused Topic Model (AAAI 2023 · similarity 0.84)
- Document Informed Neural Autoregressive Topic Models with Distributional Prior (AAAI 2019 · similarity 0.83)
- Representing Mixtures of Word Embeddings with Mixtures of Topic Embeddings (ICLR 2022 · similarity 0.83)
- Improving Topic Modeling by Distilling Soft Labels from Language Models (ICML 2026 · similarity 0.83)
- Scalable In-context Ranking with Generative Models (NeurIPS 2025 · similarity 0.83)
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.
