DeepGlobe

我们观察到，与传统图像相比，卫星图像是强大的信息来源，因为它包含更结构化和更统一的数据。尽管计算机视觉社区一直在使用深度学习来完成日常图像数据集上的艰巨任务，但卫星图像直到最近才引起人们对地图和人口分析的关注。该研讨会旨在汇集各种研究人员，以推动卫星图像分析的最新技术。

为了更多地关注此类方法，我们提出了围绕三种不同卫星图像理解任务的DeepGlobe卫星图像理解挑战。为此竞赛创建和发布的数据集可以作为卫星图像分析未来研究的参考基准。此外，由于挑战任务将涉及经典计算机视觉问题的 “野外” 形式，因此这些数据集有可能成为遥感领域以外的强大视觉算法设计的有价值的测试平台。

We observe that satellite imagery is a powerful source of information compared to traditional images, as it contains more structured and uniform data. While the computer vision community has long used deep learning to tackle challenging tasks on everyday image datasets, satellite imagery only recently garnered attention for map-making and population analysis. This workshop aims to bring together researchers from various backgrounds to advance the state-of-the-art in satellite image analysis.

To further focus attention on such approaches, we present the DeepGlobe Satellite Image Understanding Challenge, which centers on three distinct satellite image understanding tasks. The datasets created and released for this competition can serve as reference benchmarks for future research in satellite image analysis. Furthermore, as the challenge tasks will involve "in-the-wild" variants of classic computer vision problems, these datasets have the potential to become valuable testbeds for developing robust visual algorithms beyond the remote sensing domain.