Research map / Graph learning / Graph self-supervised learning
Sat2Graph: Road Graph Extraction through Graph-Tensor Encoding
Songtao He, Favyen Bastani, Satvat Jagwani, Mohammad Alizadeh, Hari Balakrishnan, Sanjay Chawla, Mohamed M. Elshrif, Samuel Madden, Mohammad Amin Sadeghi
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 graph-tensor encoding (Sat2Graph-DLA vs Seg-DLA, same DLA backbone) (ablation) | DLA segmentation model · City-Scale dataset | TOPO F1: 73.89 -> 76.26 | Improved |
| add graph-tensor encoding (Sat2Graph-DLA vs Seg-DLA, same DLA backbone) (ablation) | DLA segmentation model · City-Scale dataset | APLS: 57.22 -> 63.14 | Improved |
| add graph-tensor encoding (Sat2Graph-DLA vs Seg-DLA, same DLA backbone) (ablation) | DLA segmentation model · SpaceNet Roads dataset | TOPO F1: 74.11 -> 80.97 | Improved |
| add graph-tensor encoding (Sat2Graph-DLA vs Seg-DLA, same DLA backbone) (ablation) | DLA segmentation model · SpaceNet Roads dataset | APLS: 56.36 -> 64.43 | Improved |
| use GTE with the U-Net backbone (ablation) | Seg-UNet · City-Scale dataset | TOPO F1: 70.36% to 76.40% | Improved |
| use GTE with the Seg-DRM backbone (ablation) | Seg-DRM · City-Scale dataset | TOPO F1: 73.80% to 74.66% | Improved |
| vary cosine-distance weight w in decoding from 1 to 25 (ablation) | Sat2Graph-DLA · City-Scale dataset | TOPO F1: 75.87% -> 76.72% | Improved |
| raise w from 25 to 150 (ablation) | Sat2Graph-DLA · City-Scale dataset | TOPO F1: 76.72% -> 75.68% | Hurt |
| vary w in the range 5 to 100 (ablation) | Sat2Graph-DLA · City-Scale dataset | TOPO F1: greater than 76.2% | No clear effect |
| use a higher edge probability threshold than the vertex threshold (ablation) | Sat2Graph · City-Scale dataset | TOPO | Improved |
Later papers that compared with this method
2 later accepted papers report Sat2Graph as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- TD-Road: Top-Down Road Network Extraction with Holistic Graph Construction (ECCV 2022 · better on 5 of 5 shared results)
- Relationformer: A Unified Framework for Image-to-Graph Generation (ECCV 2022 · 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
- LineGraph2Road: Structural Graph Reasoning on Line Graphs for Road Network Extraction (ECCV 2026 · similarity 0.91)
- Patched Line Segment Learning for Vector Road Mapping (AAAI 2024 · similarity 0.90)
- Towards Satellite Image Road Graph Extraction: A Global-Scale Dataset and A Novel Method (CVPR 2025 · similarity 0.88)
- Mastering Spatial Graph Prediction of Road Networks (ICCV 2023 · similarity 0.88)
- VecRoad: Point-Based Iterative Graph Exploration for Road Graphs Extraction (CVPR 2020 · similarity 0.88)
- Road Network Representation Learning with the Third Law of Geography (NeurIPS 2024 · similarity 0.85)
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.
