And again an interesting project, which manipulates faces with the help of Nvidias Deep Learning algorithm StyleGAN: SAM (Style-based Age Manipulaton) of an Israeli research team of the University of Tel Aviv is specialized on it, starting from only one photo the face on it to change purposefully the age, to make it younger or also older.
The special feature of the new method: the age-related changes in the appearance of a face happen independently of other factors, i.e. only the age changes but not also, for example, the expression of a face, hairstyle, head forum or the direction of the eyes - this was not possible before. Also, the method works for special challenges such as a side view of a face including sunglasses.
Subsequent targeted modification of individual features of an altered face, such as facial expression, facial hair or hairstyle, is also easy - so different versions of an aged face can be easily produced - with or without a beard, graying or not.
The results look very good, but the algorithm is still far from perfect. However, this ability is bought by a shortcoming - the generated age variations of a face look realistic, but the resemblance - compared to a real photo - is not always very accurate.
But as always in the extremely fast developing field of AI: once the first step is made and a method works basically in a relatively short time ("two more papers down the line" or two studies later) still existing problems or limitations are solved. This is usually also true for the application of such algorithms to moving images - first a method works for photos and a little later it is extended to videos (i.e. chronological image sequences).
Here is the Two Minute Papers presentation of the SAM algorithm:
But the nice thing is that thanks to Replicate.com everybody can try the new method with an own photo by himself - necessary is only a registration on GitHub to avoid mass abuse, because the computational effort per image is quite high.