Large Motion Video Super-Resolution with Dual Subnet and Multi-Stage Communicated Upsampling

Video super-resolution (VSR) aims at restoring a video in low-resolution (LR) and improving it to higher-resolution (HR). Due the characteristics of tasks, is very important that motion information among frames should be well concerned, summarized utilized for guidance VSR algorithm. Especially, when contains large motion, conventional methods easily bring incoherent results or artifacts. In this paper, we propose novel deep neural network with Dual Subnet Multi-stage Communicated Upsampling (DSMC) videos motion. We design new module named U-shaped residual dense 3D convolution (U3D-RDN) fine implicit estimation compensation (MEMC) as coarse spatial feature extraction. And present Multi-Stage (MSCU) make full use intermediate upsampling guiding VSR. Moreover, dual subnet devised aid training our DSMC, whose loss helps reduce solution space enhance generalization ability. Our experimental confirm method achieves superior performance on compared state-of-the-art methods.