Synthinel-1

Synthinel-1是由杜克大学电气与计算机工程系开发的合成高分辨率卫星图像数据集，专为建筑物分割任务设计。该数据集包含2108张572×572像素的图像，分辨率为0.3m/pixel，涵盖九种不同的虚拟城市风格。数据集的创建过程利用了CityEngine软件，快速生成大规模随机化的虚拟环境，无需手动标注即可获取图像。Synthinel-1主要应用于深度学习模型的训练，特别是在处理来自新地理区域或条件下的卫星图像时，能够显著提升模型的泛化能力和性能。

Synthinel-1 is a synthetic high-resolution satellite image dataset developed by the Department of Electrical and Computer Engineering at Duke University, specifically designed for building segmentation tasks. This dataset contains 2108 images with a size of 572×572 pixels and a spatial resolution of 0.3 meters per pixel, covering nine distinct virtual urban styles. The dataset was created using CityEngine software, which enables rapid generation of large-scale randomized virtual environments, allowing automatic acquisition of images without manual annotation. Synthinel-1 is primarily used for training deep learning models, and can significantly improve the generalization ability and performance of the models, especially when processing satellite images from new geographic regions or under different conditions.