Researcb On Higb-Qnabty Arbitrary Humau Posc Image And Video Generation

Image and video generation is a synthetic image/video technology that converts the input source person into any given target poses,whose appearance and the texture are consistent with the input image.Image and video generation can be applicable to a wide range of scenes,such as interactive entertainment,film special effects synthesis,fashion modeling design and image/video data set enhancement.However,current generation methods usually ignore detailed appearance and texture information of human images in image generation technology,and models are complex and difficult to train.Existing video generation algorithms neglect the influence of the cinematic background environment on the accuracy of human pose estimation,and the combination of backgrounds in the field of motion transfer research lacks visual appeal.For the above problems,this article conducts research on image and video generation technology,the main contents and innovations include:1.In the field of image generation,multi-scale conditional generation adversarial networks consisting of two generators and two discriminators are proposed to solve the problem of producing person image with precise pose and preserve appearance details simultaneously.The first generator transforms the conditional person image into a coarse image of the target pose globally,and the other is to enhance the detailed quality of the synthetic person image through a local reinforcement network.The outputs of the two generators are then merged into a synthetic,discriminant and high-resolution image.Joint global generation and local refinement can model both the accuracy and quality of the synthetic image simultaneously.Experiments are conducted on the Market-1501 and DeepFashion datasets to evaluate the proposed model,and both qualitative and quantitative results demonstrate the superior performance of the proposed model.2.In the field of video generation,a pose-guided scene-preserving person video generation algorithm is proposed.First,the video frame with the appearance of the segmented character is used as the network input instead of the source video frame;Then,a motion transformation model is employed to replace characters in source video with target characters,and keep the consistency of motion.Finally,the character appearance is fused with the background of the source character,which has flowed advantages;a)removing border anomaly pixels;b)realizing character blending naturally into the source scene;and c)avoiding changing the background environment and overall image style.The human skeleton diagram is selected as the intermediate representation of the source video and the target video to reduce background interference,and improve the accuracy of pose estimation,which can naturally realize scene-preserving during the motion transfer process,and produce person video that is both artistic and authentic.3.Pose-guided scene-preserving person video generation algorithm is the first method to achieve target pose transfer and background switching between different video character objects.It can complete the motion transfer and background conversion of the source character and the target character.Demonstrate that the target person performs the actions of the source character in the source background.Compared with other fusion methods,this method completely integrates the target person and the source background environment without obvious boundary pixel difference,which has significant application prospects and economic value.