Research map / Architectures and efficiency / RNNs and normalization
Beyond the Pixel-Wise Loss for Topology-Aware Delineation
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 |
|---|---|---|---|
| Add VGG19-feature topology loss to pixel-wise BCE (ablation) | U-Net · Cracks | Quality: 0.3924 -> 0.7446 | Helped |
| Add VGG19-feature topology loss to pixel-wise BCE (ablation) | U-Net · Roads | Quality: 0.5152 -> 0.5719 | Helped |
| Add VGG19-feature topology loss to pixel-wise BCE (ablation) | U-Net · EM | Quality: 0.5406 -> 0.5580 | Helped |
| Add iterative refinement to topology-loss training (ablation) | OURN · Cracks | Quality: 0.7446 -> 0.8461 | Helped |
| Add iterative refinement to topology-loss training (ablation) | OURN · Roads | Quality: 0.5719 -> 0.6524 | Helped |
| Add iterative refinement to topology-loss training (ablation) | OURN · EM | Quality: 0.5580 -> 0.5722 | Helped |
| Use VGG19 layers 1, 2, and 3 rather than no VGG features (ablation) | OURN · Cracks | Quality: 0.4050 -> 0.7151 | Helped |
| Use VGG19 layers 1, 2, and 3 rather than layer 3 alone (ablation) | OURN · Cracks | Quality: 0.6974 -> 0.7151 | Helped |
| Increase refinement from one to three iterations (ablation) | OURS · EM | Quality: 0.5621 -> 0.5722 | Helped |
| Increase refinement from three to four iterations (ablation) | OURS · EM | Quality: 0.5722 -> 0.5727 | No clear effect |
Closest papers by meaning
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.87)
- Diffeomorphic interpolation for efficient persistence-based topological optimization (NeurIPS 2024 · similarity 0.86)
- DeepBranchTracer: A Generally-Applicable Approach to Curvilinear Structure Reconstruction Using Multi-Feature Learning (AAAI 2024 · similarity 0.86)
- Towards Scalable Topological Regularizers (ICLR 2025 · similarity 0.84)
- Generalized Linear Mode Connectivity for Transformers (NeurIPS 2025 · similarity 0.84)
- TopoDiffusionNet: A Topology-aware Diffusion Model (ICLR 2025 · similarity 0.84)
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.