Research map / Label-efficient and robust learning / Domain adaptation and generalization
Leveraging Crowdsourced GPS Data for Road Extraction From Aerial Imagery
Tao Sun, Zonglin Di, Pengyu Che, Chun Liu, Yin Wang
Check 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 GPS input to aerial imagery | D-LinkNet · Beijing test set | IoU (%): 54.42 -> 56.96 | Improved |
| Replace 3×3 transpose convolutions with a 1D decoder (ablation) | D-LinkNet with image + GPS input · Beijing test set | IoU (%): 56.96 -> 57.96 | Improved |
| Add the GPS sampling-interval channel (ablation) | D-LinkNet with image + GPS and 1D decoder · Beijing test set | IoU (%): 57.96 -> 58.55 | Improved |
| Add GPS data augmentation (ablation) | D-LinkNet with image + GPS, interval, and 1D decoder · Beijing test set | IoU (%): 58.55 -> 59.18 | Improved |
| Add GPS input to image input | D-LinkNet · Shanghai test set, trained on Beijing | IoU (%): 38.16 -> 44.57 | Improved |
| Add GPS augmentation (ablation) | Image + GPS model · Shanghai test set, trained on Beijing | IoU (%): 44.57 -> 48.69 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | U-Net with GPS input · Beijing test set | IoU (%): 43.63 -> 48.10 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | Res U-Net with image + GPS input · Beijing test set | IoU (%): 52.29 -> 54.24 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | LinkNet with image + GPS input · Beijing test set | IoU (%): 57.48 -> 57.89 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | D-LinkNet with image input · Beijing test set | IoU (%): 54.42 -> 55.15 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | Res U-Net with GPS input · Beijing test set | IoU (%): 45.33 -> 48.52 | Improved |
| Use a 1D decoder instead of a plain decoder (ablation) | LinkNet with image input · Beijing test set | IoU (%): 53.96 -> 54.84 | Improved |
Closest papers by meaning
- Beyond Road Extraction: A Dataset for Map Update Using Aerial Images (ICCV 2021 · similarity 0.87)
- DeepRoadMapper: Extracting Road Topology From Aerial Images (ICCV 2017 · similarity 0.87)
- RoadTracer: Automatic Extraction of Road Networks From Aerial Images (CVPR 2018 · similarity 0.87)
- DeepDualMapper: A Gated Fusion Network for Automatic Map Extraction Using Aerial Images and Trajectories (AAAI 2020 · similarity 0.86)
- Patched Line Segment Learning for Vector Road Mapping (AAAI 2024 · similarity 0.83)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · 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.
