Animating faces from still images
Problem being addressed
Given a still image of someone's face, can we automatically generate a video sequence of what they would look like moving their head and talking?
Given very few images of a person's headshot, this approach can automatically generate a video sequence of what the person may look like if their head is moving around or if they are talking. From a single image, a video with a reasonable level of quality can be generated. The more images that you have of a single person, the more realistic the videos generated become, with perfect realism being achieved when the technology is provided around 30 images of the same person. A deep learning approach is used.
Advantages of this solution
Very few headshot photos are required to achieve realistic results. This is an advantage compared to other approaches where you may need dozens (if not hundreds or more) of headshots of a single person before you can generate a video sequence of them.
Possible New Application of the Work
This could be used by an Advertising company to bring their headshots of models to life. From a single headshot, they could generate a video of that person's head moving. It would cost very little to generate such a video as the company would not have to bring the model back in and have them shoot the video.
Source URL: #############