Reproduce “Building and Road Recognition in Dense Urban Informal Settlements: A Dataset and Benchmark” 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
Hongyu Long, Jiaxuan Liu, Rui Cao
Researchers from HKUST(GZ) introduced the DenseUIS dataset, the first high-resolution remote sensing dataset specifically tailored for building and road extraction in dense urban informal settlements. Benchmarking state-of-the-art deep learning models on this dataset revealed that while RS-Mamba demonstrated comparatively better performance, particularly in recall and connectivity, all existing methods still face significant challenges in accurately mapping infrastructure in these complex environments.
Key insights
- The unique and complex morphology of dense urban informal settlements, with highly occluded and irregular features, poses a substantial challenge for current deep learning models, leading to issues like 'adhesion effects' for buildings and fragmentation for roads.
- Architectures capable of modeling global context and long-range dependencies, such as RS-Mamba, generally exhibit superior performance compared to traditional CNNs by better capturing continuity and boundary information in these ambiguous environments.
- Despite the advancements of modern deep learning, the absolute performance scores, particularly for road extraction (max IoU 45.92%), highlight a significant gap and the necessity for developing specialized methods tailored to the specific characteristics of informal settlements.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.