The effects of super-resolution on object detection performance in an aerial image

Abstract

Image super-resolution (SR) has a positive effect on the problem of detecting objects on low resolution (LR) images. In this study, we train custom RCAN to create SR images from LR, and at the same time train Practice common object detection methods Faster RCNN, Cascade-RCNN, DetectoRS, Retina, SSD on both SR, LR datasets. Experimental results, proving that SRx2 significantly improves subject detection results of LRx2 images. And We make a comparison between the results of SR and HR.