Video Synthesis Method Based on Cascade Refinement Network
A high-resolution video to video generation method is proposed based on the image to im-age synthesis algorithm to address the problem of poor video generation quality and inability to continue the generated object attributes in subsequent videos,resulting in a decrease in the visual effect of simulated videos.Adding residual blocks to the cascaded optimization network to optimize the network structure and improve the quality of generated video frames.In order to solve the problem that the attributes of the gen-erated objects are inconsistent in subsequent videos,the optical flow is calculated by two improved casca-ded optimization network prediction images,and then one image is predicted by optical flow.The two pre-dicted images are fused to obtain the simulation video sequence.Compared with other video and image synthesis methods on cityscapes dataset,the results show that the proposed algorithm can get more realistic video,and the generated video sequences have higher evaluation.
deep learningvideo to video synthesisimage style transferoptical flow estimation
NETL
NSTL
万方数据
