Kemelmacher-Shlizerman told .uk that the biggest challenge was coming up with a method for "completely automatic analysis of face photos 'in the wild'", i.e. The study authors say that future improvements for the work include: modelling wrinkles and hair whitening to enhance the realism of older subjects, increasing the range of ethnicities, and having a database of heads and upper torsos of different ages in order to apply the same technique to. The results seemed to show that for ageing young children, the University of Washington's technique outperformed all prior work. 37 percent (out of 8,916 votes) said that the University of Washington team's approach was more likely to be the older baby, 44 percent saying the actual image was more likely.ġ5 percent of people said that both were equally likely to be the adult version of the baby, while five percent said neither were likely. The results seem to show that humans identified the generated image as the older version almost as often as they identified the actual older image. The volunteers had to say which of the two older photos were more like the baby.
#Face morph age progression sur windows 7 software#
One picture would be an individual as a baby, and the two additional photos would be that person at a specific age (say 25) - one generated by the software and one actual image of that person at that age. This was put to the test by showing three pictures to human subjects (through Amazon's Mechanical Turk). This allowed them to see how effective the software was at accurately ageing the children. To check the efficacy of the system, the team fed in child images of individuals for whom they also had adolescent and adult images.
0 kommentar(er)
