Motion Transfer & Person Image Synthesis

Human motion transfer from a video to an image is one of the applications of GAN. This paper includes the techniques and GAN Models which transfers motion from a video to an image and produces digitally synthesized video of the person present in the image. This paper provides the insight of the Nested GAN model. It generates the human images along with the human features, pose, garments, etc., that is present in the wide source of images. The proposed method consists of two stages. First, the motion patterns are extracted from driving video sequences. Then image-to-image transformations are learned by maximizing the mutual information between the source image and the target image while simultaneously minimizing the distortion associated with the image-to-image transformations. Simulation results show that using our method reduces work factor values, and generates higher quality videos. Method discussed in this paper enables the generation of more realistic output images as well as provides the separated un-annotated attributes of the images. The suggested method along with some innovation outperforms the latest pose transfer and image synthesis models according to experimental results.