☨ Work partially done while interning at MPI for Informatics
* Equal senior contribution
The following shows reconstructed videos re-rendered under a set of target styles. Each target style is composed of style codes for static and for playable objects. In each column we show a target video from which the style is obtained. In each row we show the source video which is re-rendered under each target style.
On the Minecraft dataset, our model correctly alters player appearances and changes the illumination conditions according to the target style.
On the Tennis dataset, the model correctly alters the appearance of the field and of the players. Note that the motion of each player is not altered, showing the disentanglement of style and pose.