There has been an explosion of research into computer vision focused on deep learning. These significant advances in image classification, object detection and image segmentation have profound implications for foundational mapping applications. Recent open source initiatives such as SpaceNet have strived to direct more research and development towards remote sensing applications. GIS practitioners need to understand and engage the research community to help structure the application of these new techniques against geospatial problems. tt is difficult translate mission requirements to machine learning evaluation metrics, and vice versa. For example, in the computer vision community, most results are described by certain image specific metrics such as mAP, F1Score, Precision and Recall. Alternatively, a GIS practitioner may want to incorporate machine learning capabilities into their workflow, but not know what level of performance is necessary for the specific mission. We will discuss a framework for defining levels of practitioners augmentation that will allow end-user groups and machine learning researchers to better understand each other and help direct the application of these advanced algorithms against geospatial problems.
None