Deep Learning Based General Object Detection Research And Application

Object detection is an important task in computer vision.Most current detection methods have adopted anchor boxes as regression references.However,the detection performance is sensitive to the setting of the anchor boxes.When employing some approaches on different tasks or datasets,it is usually indispensable to redesign the anchor boxes.A proper setting of anchor boxes may vary significantly across different datasets,which severely limits the universality of the detectors.To improve the adaptivity of the detectors,in this paper,we present a novel dimension-decomposition region proposal network(De RPN)that can perfectly displace the traditional Region Proposal Network(RPN).De RPN utilizes an anchor string mechanism to independently match object widths and heights,which is conducive to treating variant object shapes.In addition,a novel scale-sensitive loss is designed to address the imbalanced loss computations of different scaled objects,which can avoid the small objects being overwhelmed by larger ones.We conduct a series of comprehensive experiments on some public datasets for general object detection,including PASCAL VOC 2007,PASCAL VOC 2012 and MS COCO.We evaluate the region proposals generated by De RPN and RPN,respectively.Also,we verify the overall improvements from the De RPN.All these experiments prove that our De RPN can significantly outperform RPN.Besides,we apply the proposed De RPN to scene text detection to further verify its adaptivity.Without any specialized optimization,De RPN can achieve good performances on both ICDAR 2013 and COCO-Text datasets,and it can even outperform some specific scene text detectors.To sum up,the proposed De RPN can be employed directly on different models,tasks,and datasets without any modifications of hyperparameters or specialized optimization.De RPN has terrific adaptivity.