A single road segmentation model, trained with cross-resolution distillation and topology losses across a resolution curriculum, outperforms prior methods on global and city benchmarks without per-domain fine-tuning.
arXiv preprint arXiv:2410.22629 (2024) 3
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Distilled Roads: Generalisable Road Network Extraction Across Sensors, Resolutions, and Region
A single road segmentation model, trained with cross-resolution distillation and topology losses across a resolution curriculum, outperforms prior methods on global and city benchmarks without per-domain fine-tuning.