Reproduce “Towards Satellite Image Road Graph Extraction: A Global-Scale Dataset and A Novel Method” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Pan Yin, Kaiyu Li, Xiangyong Cao, Jing Yao, Lei Liu, Xueru Bai, Feng Zhou, Deyu Meng
Researchers from Xi'an Jiaotong University, Chinese Academy of Sciences, and Xidian University introduced Global-Scale, a 20x larger and more diverse dataset for road graph extraction, and SAM-Road++, a method that improves accuracy and generalization. SAM-Road++ achieved higher F1 scores on both the novel Global-Scale dataset (62.33 F1 in-domain, 48.34 F1 out-of-domain) and existing benchmarks compared to prior state-of-the-art.
Key insights
- A large-scale and geographically diverse dataset is essential for training robust road graph extraction models that can generalize effectively to unseen real-world scenarios.
- Addressing discrepancies between training and inference inputs, specifically for node positions in two-stage graph extraction models, leads to improved topological accuracy and overall performance.
- Simple yet effective strategies that leverage contextual clues, such as extended road lines, can significantly enhance a model's ability to maintain connectivity in the presence of occlusions.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.