Project Leader
Prof. Dr. Xin Sun (Humboldt Research Fellowship for experienced researchers)

Host institute
Prof. Dr. Xiaoxiang Zhu
Chair of Data Science in Earth Observtion, Technische Universität München

Cooperation Partners
Ocean University of China, University of Vienna 

The past decade has witnessed major advances in deep learning for object detection in vision tasks. However, the flourish of deep learning is mainly accredited to the rich annotated data. The gaps between experimental environment and realistic scenarios result in catastrophic performance degradation when the algorithms are deployed in practical satellite images. This project focuses on the object detection in the domain specific scenarios, which faces the noteworthy and typical real-world applications with environmental constraints. Two main issues in this research field are eager to be tackled. The first one is how to extract robust deep features with insufficient and partially annotated training set, i.e., weakly supervised deep learning for object detection on satellite image. The second is how to detect objects of novel categories with only a few given examples, i.e., few-shot object detection.