Research map / Learning theory and algorithms / Prediction and decision losses
Promoting Connectivity of Network-Like Structures by Enforcing Region Separation
Doruk Öner, Mateusz Koziński, Leonardo Citraro, Nathan C. Dadap, Alexandra G. Konings, Pascal Fua
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 |
|---|---|---|---|
| add windowed TOPO loss to U-Net | U-Net · RoadTracer | APLS: 66.3 -> 75.8 | Improved |
| add windowed TOPO loss to DRU | DRU · DeepGlobe | APLS: 75.2 -> 77.1 | Improved |
| add windowed TOPO loss to DRU | DRU · Canals | TLTS: 32.7 -> 43.1 | Improved |
| add windowed TOPO loss to U-Net | U-Net · Canals | APLS: 70.9 -> 76.3 | Improved |
| compute loss in 64x64 windows instead of globally (ablation) | U-Net+TOPO · RoadTracer | APLS: 72.5 -> 75.8 | Improved |
| compute loss in windows instead of globally (ablation) | U-Net+TOPO · RoadTracer | JCT: 84.7 -> 82.8 | Hurt |
| set alpha to 1e-4 instead of 0 (add topology term) (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 66.3 -> 75.8 | Improved |
| raise alpha from 1e-4 to 1e-3 (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 75.8 -> 72.3 | Hurt |
| lower alpha from 1e-4 to 1e-5 (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 75.8 -> 71.4 | Hurt |
| use beta 0.1 instead of 1 for the false-connection term (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 71.4 -> 75.8 | Improved |
| use beta 0.1 instead of 0.01 (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 74.3 -> 75.8 | Improved |
| increase window size from 16x16 to 64x64 (ablation) | U-Net+TOPO-windowed · RoadTracer | APLS: 68.3 -> 75.8 | Improved |
Closest papers by meaning
- RoadTracer: Automatic Extraction of Road Networks From Aerial Images (CVPR 2018 · similarity 0.89)
- TopoAL: An Adversarial Learning Approach for Topology-Aware Road Segmentation (ECCV 2020 · similarity 0.87)
- Improved Road Connectivity by Joint Learning of Orientation and Segmentation (CVPR 2019 · similarity 0.87)
- Reconstructing Curvilinear Networks Using Path Classifiers and Integer Programming (TPAMI 2016 · similarity 0.87)
- Towards Reliable Evaluation of Algorithms for Road Network Reconstruction from Aerial Images (ECCV 2020 · similarity 0.85)
- Patched Line Segment Learning for Vector Road Mapping (AAAI 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.
