| Abstract [eng] |
Image animation is a process in which, image is animated based on motion of another video with an object of the same category. This is achieved by training a neural network to reconstruct video, using the first frame of the video and motion representations. Neural network models presented in research papers are trained using as simple data, as possible, in which an object is always filmed from the same location and is performing simple animations. This allows to achieve photorealistic results, but practical application remains very narrow. In this paper, it will be showed, how image animation can be applied in a complex video environment in which: objects are different in appearance, different motions are performed, object distance between cameras varies and objects are at different locations in frame. Successful image animation application, would allow to photorealistically edit the video by changing an object's appearance. |
