Research map / Detection and segmentation / CNN features
Hilbert Curve-Based Attention Enabling Topology-Preserving Image Tensor Representation for Semantic Segmentation Network
Linkang Xu, Gang Li, Yue Song, Xiangxin Ji
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 |
|---|---|---|---|
| flatten with Hilbert curve instead of row-first traversal in the decoder attention (ablation) | TPDecoder (Swin-T) · BD3 building defects | mIoU: 75.58 -> 79.49 | Improved |
| flatten with Hilbert curve instead of Z-order traversal (ablation) | TPDecoder (Swin-T) · BD3 building defects | mIoU: 77.89 -> 79.49 | Improved |
| flatten with Hilbert curve instead of row-first traversal (ablation) | TPDecoder (Swin-T) · BD3 building defects | ACC: 86.37 -> 89.51 | Improved |
| use Hilbert curve instead of row-first or Z-order for flattening (ablation) | 2D flattening · locality loss, order 7 | locality loss: 2064766 (Row-First) -> 16383 (Hilbert) | Improved |
| use Hilbert curve instead of row-first flattening (ablation) | 2D flattening · runtime, 128x128 | runtime (s): 0.009974 -> 0.130650 | Hurt |
| add Joint Auxiliary Loss (Dice + edge-aware) with weights dice 0.5, edge 1.0 (ablation) | TPSegformer (Swin-T) · BD3 building defects | mIoU: 79.49 -> 80.77 | Improved |
| add Joint Auxiliary Loss with weights dice 0.5, edge 1.0 (ablation) | TPSegformer (Swin-T) · BD3 building defects | ACC: 89.51 -> 90.22 | Improved |
| scale Swin backbone from Tiny to Large (ablation) | TPDecoder · BD3 building defects | mIoU: 79.49 -> 81.37 | Improved |
| scale Swin backbone from Tiny to Base (ablation) | TPDecoder · BD3 building defects | ACC: 89.51 -> 88.57 | Hurt |
| scale ResNet backbone from 50 to 101 layers (ablation) | TPDecoder · BD3 building defects | mIoU: 79.00 -> 78.22 | Hurt |
| use TPSegformer instead of the best baseline | segmentation network · Dacl10k | mIoU: 36.21 -> 44.27 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SegFormer — SegFormer: A Topic Segmentation Model with Controllable Range of Attention (AAAI 2023) · better on 2 of 2 shared results
- CCNet — CCNet: Criss-Cross Attention for Semantic Segmentation (ICCV 2019) · better on 2 of 2 shared results
- BiSeNetV2 — BiSeNet V2: Bilateral Network with Guided Aggregation for Real-Time Semantic Segmentation (IJCV 2021) · better on 1 of 1 shared result
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
- Joint Topology-Preserving and Feature-Refinement Network for Curvilinear Structure Segmentation (ICCV 2021 · similarity 0.85)
- TopoRefine: Plug-and-Play Topology-Aware Contour Refinement for Building Segmentation (NeurIPS 2026 · similarity 0.85)
- SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformers (NeurIPS 2021 · similarity 0.84)
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.83)
- Rethinking Decoders for Transformer-based Semantic Segmentation: A Compression Perspective (NeurIPS 2024 · similarity 0.83)
- Geometric Exploitation for Indoor Panoramic Semantic Segmentation (NeurIPS 2024 · similarity 0.82)
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.
