I. Introduction
Augmented reality (AR) has received significant attention across the industry and academic society [1]–[4]. Recently, AR technologies pursue methods for more realistic experience, which has also been studied across various domains [5]–[11]. To meet the demand for more realistic AR applications, developers started to pursue higher level user interactions. For realistic experience in such interactions, recognizing the surrounding environment in real-time is highly necessary. A suitable technology to satisfy this requirement is object detection. Object detection can be considered the most basic form of environment recognition since knowing which object is present in the current scene can provide plenty information about the environment. Also, object detection have shown astonishing development [12]–[19] and now one stream of the detection systems guarantee real-time (>30 fps) performance with high reliability [17]–[19]. By customizing these object detection systems, we can provide better experience reliably in real-time for AR applications.