Research map / Detection and segmentation / CNN features
Structure-Aware Image Segmentation with Homotopy Warping
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 homotopy warping loss to UNet | UNet · RoadTracer | Betti error: 1.591 -> 1.251 | Helped |
| add homotopy warping loss to 3D UNet | 3D UNet · CREMI | DICE: 0.961 -> 0.967 (DMT 0.973) | Mixed |
| warping loss vs DMT on Mass. ARI | Homotopy Warping · Mass. | ARI: 0.881 (DMT) vs 0.864 | Hurt |
| use distance-transform ordering when warping (ablation) | Homotopy Warping · RoadTracer | Betti error: 1.473 -> 1.251 | Helped |
| warp in both directions instead of GT->Pred only (ablation) | Homotopy Warping · RoadTracer | ARI: 0.567 -> 0.572 | Helped |
| use cross-entropy instead of MSE as pixel loss in L_warp (ablation) | Homotopy Warping · RoadTracer | Betti error: 1.429 -> 1.251 | Helped |
| use cross-entropy instead of Dice as pixel loss in L_warp (ablation) | Homotopy Warping · RoadTracer | DICE: 0.606 (Dice loss) vs 0.603 (CE) | Hurt |
| vary lambda_warp from 2e-5 to 2e-4 (ablation) | Homotopy Warping · RoadTracer | Betti error: best 1.251 at 1e-4; 1.412 at 5e-5 | Mixed |
| add warping loss instead of morphological closing post-processing (ablation) | UNet · RoadTracer | Betti error: 1.583 -> 1.251 | Helped |
| apply morphological closing post-processing to UNet (ablation) | UNet · RoadTracer | Betti error: 1.591 -> 1.590 | No clear effect |
Closest papers by meaning
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.92)
- SegGraph: Leveraging Graphs of SAM Segments for Few-Shot 3D Part Segmentation (NeurIPS 2025 · similarity 0.83)
- TopoDiffusionNet: A Topology-aware Diffusion Model (ICLR 2025 · similarity 0.83)
- Complex Wavelet Mutual Information Loss: A Multi-Scale Loss Function for Semantic Segmentation (ICML 2025 · similarity 0.83)
- Efficient Connectivity-Preserving Instance Segmentation with Supervoxel-Based Loss Function (AAAI 2025 · similarity 0.83)
- Scale-Free Image Keypoints Using Differentiable Persistent Homology (ICML 2024 · 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.