Yingguan Wang

1. Introduction In recent decades, one-shot or few-shot detection (OSD and FSD for short, respectively) and localization have seen broad application prospects. Currently, most OSD or FSD methods rely heavily on pre-training, which usually involves mass data and a large amount of computational resources.

1. Introduction One-shot or few-shot object detection (OSD and FSD, respectively) and localization have been widely demanded in visual-related projects. Specifically, in our project, which has hardly any training examples, in a scenario where a specified target is set to mark terrains, a model that can not only perform one-shot object detection on simple targets but can also output their location is needed.