Research map / Detection and segmentation / CNN features
T-measure: A Topology-Consistent Metric for Binary Segmentation
pengfei zhang, Jian 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 |
|---|---|---|---|
| use Tm instead of Sm to score empty predictions on small objects | Sm · ISIC2017 small objects (153) | average score on empty prediction: 0.482 -> 0.00 | Improved |
| use Tm instead of Sm to detect fragmented masks | Sm · COD10K (200 pairs) | relative score drop on fragmented masks: -11.6% -> -84.3% | Improved |
| add topological integrity term to metric (Tm vs pixel metrics) on fracture test | IoU, Dice, Fm, Em, MAE · Controlled P1 vs P2 example | score difference between P1 and P2: 0.0% -> -30.4% | Improved |
| use boundary-distance metrics 95HD or ASD for fracture detection | 95HD, ASD · Controlled P1 vs P2 example | score of fractured prediction: 95HD 27.00 -> 12.65; ASD 2.93 -> 1.26 | Hurt |
| use Tm instead of Sm for agreement with LPIPS object extraction on DUTS | Sm · DUTS | Spearman rho with LPIPS: 0.610 -> 0.779 | Improved |
| use Tm instead of pixel-based MAE or weighted F on LPIPS consistency | MAE, F-weighted · DUTS, COD10K, CrackSeg9k | Spearman rho with LPIPS: DUTS 0.792 -> 0.779 | No clear effect |
| choose Tm's preferred map over Sm's preferred map in conflicting pairs | Sm · 50 conflicting pairs from DUTS and COD10K | human preference rate: 70.7% vs 50% chance | Improved |
| add neighborhood weighting (weighted F-measure) instead of plain pixel metrics | F-weighted · Controlled P1 vs P2 example | score of fractured prediction: 0.942 -> 0.943 | Hurt |
Closest papers by meaning
- Topograph: An Efficient Graph-Based Framework for Strictly Topology Preserving Image Segmentation (ICLR 2025 · similarity 0.88)
- Topologically Faithful Image Segmentation via Induced Matching of Persistence Barcodes (ICML 2023 · similarity 0.87)
- Topology-Aware Uncertainty for Image Segmentation (NeurIPS 2023 · similarity 0.87)
- Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation (TPAMI 2026 · similarity 0.86)
- Topology-Aware Segmentation Using Discrete Morse Theory (ICLR 2021 · similarity 0.86)
- 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.
