GAMUS
收藏资源简介:
GAMUS数据集是由慕尼黑工业大学数据科学地球观测主席团队创建的一个大型多模态语义分割基准数据集,专注于遥感数据。该数据集包含11,507个来自五个不同城市的图像块,每个图像块包含RGB图像和归一化数字表面模型(nDSM),以及像素级的语义标签。GAMUS数据集的创建旨在解决现有遥感数据集规模较小、多样性有限以及缺乏统一基准的问题,通过提供一个大规模、多样化的数据集,促进多模态学习方法的发展,特别是在RGB-Height(RGB-H)数据上的应用。数据集的创建过程涉及从公开数据源收集和处理高分辨率正射影像、语义地图和nDSM,确保数据的质量和适用性。GAMUS数据集的应用领域广泛,包括城市规划、环境监测和灾害响应等,旨在通过精确的语义分割提高对地表覆盖类型的识别和理解。
The GAMUS dataset is a large-scale multimodal semantic segmentation benchmark dataset dedicated to remote sensing data, developed by the team of the Chair of Data Science for Earth Observation at the Technical University of Munich. It comprises 11,507 image patches sourced from five distinct cities, with each patch containing RGB images, normalized digital surface models (nDSM), and pixel-level semantic labels. The creation of the GAMUS dataset aims to resolve the limitations of existing remote sensing datasets, namely their small scale, restricted diversity, and lack of unified benchmarks. By offering this large-scale and diverse dataset, it seeks to facilitate the advancement of multimodal learning methodologies, particularly their applications on RGB-Height (RGB-H) data. The dataset construction process entails collecting and processing high-resolution orthophotos, semantic maps, and nDSM from public data sources to guarantee data quality and applicability. The GAMUS dataset has a wide range of application scenarios including urban planning, environmental monitoring, disaster response and more. It aims to improve the identification and understanding of land cover types through accurate semantic segmentation.

- 1GAMUS: A Geometry-aware Multi-modal Semantic Segmentation Benchmark for Remote Sensing Data慕尼黑工业大学数据科学地球观测主席 · 2023年



