Design And Implementation Of Dance Generation System Based On Generative Adversarial Networks

Image generation aims to learn the mapping relationship between source image and target image through computer algorithm,and belongs to an important branch of image content generation in computer vision.It is currently widely used in interactive entertainment applications,image and video generation,image coloring,and reconstruction of objects based on conditions.On the other hand,in recent years,with the booming development of the anti-network,and compared with other deep learning methods,the research on generating anti-network in image content generation has more advantages.Therefore,based on the generation of anti-network technology,this thesis studies the algorithm of dance generation in the field of image generation and applies it to practical projects,which has extremely important academic significance and practical value.In practical applications,due to the flexible movement of the limbs in the dance and the change of the camera angle,the visual information of various parts of the human body changes greatly,which poses a great challenge for generating high-resolution target dance images.In order to generate a more visually appealing synthetic dance image,a dance generation algorithm based on generating a confrontational network is proposed and applied to the high resolution video generation task.According to the dance source video of a given professional dancer and the video of the target character doing a series of standard actions,the generated confrontation network technology is used to transfer the professional dance performance to the target character,and as an image with smooth time and space to Image conversion.The human skeleton map is used as an intermediate representation between the source video and the target video,so the problem is transformed into a mapping from the human skeleton to the appearance of the target subject.At the same time,the PyTorch framework and the PyQT5 framework are used to create a dance generation algorithm model and a system graphical interface design to implement the dance generation application.Two standard picture quality assessment indicators,SSIM,FID,were used to evaluate the algorithm.The experimental results ensure that the quality of the generated results also shows the effectiveness of complex dance generation,and it is also robust to the figure,scene environment and other influences,and has achieved good results.