Research And Implementation Of Video Generation System Based On Generative Adversarial Network

Image generation is to learn the mapping relationship between the source image and the target image through a computer algorithm.This technology is an important research direction in the field of image content generation in computer vision.As an important carrier of visual transmission,video contains more information than pictures,so the problem of video generation has gradually become a research hotspot.At the same time,video generation also has a wide range of application scenarios in the field of film and television entertainment.On the other hand,with the rapid development of generative adversarial networks in recent years,compared with other deep learning methods,the research of generative adversarial networks in image content generation is obviously more advantageous.Therefore,this paper will use generative adversarial network technology to solve the problem of video generation in the field of image generation,and conduct algorithm research and finally apply this technology to practical projects significance.In actual applications,the video generation methods are different.It is an important idea in video generation to transfer the motion sequence in the video to the image object through the input static image and the target object video that performs a series of standard actions.However,most of the models that implement related functions are designed for specific fields,such as video generation of human poses,and the related performance of these methods needs to be improved.Therefore,this paper proposes a video generation model based on generative adversarial networks.Unlike other video generation models,this model does not target specific research objects and can generate high-quality videos.This model uses an unsupervised key point detection module,extracts motion information through optical flow estimation,and generates realistic video images by generating an adversarial model.At the same time,the PyTorch deep learning framework is used to construct the video generation algorithm model,and the Django framework is used to build the system's graphical interface,and finally the target object's video generation system application is realized.All data for this study are from published data sets.When evaluating this algorithm,the strategy of separately evaluating motion information and content information is adopted.The average distance D_kpof key points is used to evaluate the transmission effect of motion information on the model.At the same time,when evaluating the image quality of the generated frames,two standard picture quality evaluation indicators of SSIM and FID are used.Finally,based on the comparative analysis of the experimental results,the video generation model proposed in this paper can well maintain the identity information of the object,and at the same time produce realistic visual effects,and the generated background image is consistent in time and space.The performance of the video generation model based on generative confrontation network constructed in this paper is better than the existing video generation models.