Clickable Object Detection Network for a Wide Range of Mobile Screen Resolutions

Recently, as the development cycle of applications has been shortened, it is important to develop rapid and accurate application testing technology. Since requires a lot cost, mobile component detection technology using deep learning essential prevent use expensive human resources. In this paper, we shall propose Clickable Object Detection Network (CODNet) for in wide range screen resolutions. CODNet consists three modules: feature extraction, deconvolution prediction modules order provide performance improvement scalability. The Feature Extraction module uses squeeze excitation blocks efficiently extract features change ratio input image 1:2 most close that screen. Deconvolution provides map various sizes by upsampling through top-down pathway lateral connections. selects an anchor size suitable environment Anchor Transfer block, among set candidates obtained analysis dataset. Moreover, improve object building new dataset consisting data collected from resolutions operating systems. We show our model achieves competitive mean average precision on compared other models.