Artificial intelligence turns pixelose into realistic portrait photos
Face Depixelizer is a new, supported by artificial intelligence, application that can convert a small pixelated image into a realistic high-resolution portrait photo. It is impressive, but it also has its drawbacks.
The creator of the Face Depixelizer application is Russian Denis Malimonov. The base of his app is StyleGAN - Generative Competitive Network (GAN) made available by Nvidia researchers in December 2018.
StyleGAN may be associated with thispersondoesnotexist.com . A collection of realistic portrait photos of non-existent people by Phillip Wang, a developer of Uber.
Face Depixelizer can generate a realistic high-resolution portrait picture.
All he needs is a pixelated small picture on which nothing can be seen. The whole process is as follows:
After adding the input image, the algorithm generates the mass of different faces using the StyleGAN program and pixelates them, compared to the original image. When he finds the perfect pair, it means that the face in the image generated by Face Depixelizer may be similar to the real one. It may or may not.
The problem is that pixelation of photos is very lossy for the original image. All details and characteristics disappear. This means that different faces may look identical when reduced and pixelated.
The creator emphasizes that his application cannot perfectly reproduce the image of the pixelated person in the photo
Some who look at the possibilities of such systems come to mind scenes from movies :
The "Enhance" function has passed into the canons of film and has been used for years in the whole mass of production. She even got memes about herself.
When I watched as a child, McGiver, Fox Mulder or Captain Jean-Luc Picard asked to "enlarge and sharpen the image," I was perfectly aware that it is fiction and in fact machines cannot do such things. In a few years, children looking at similar scenes will understand it a bit differently.
However, we can imagine that with time and the development of technology, this system can gain such potential as in movies. It is already able to help outline the appearance of the face, which can be useful to create images of sought after people based on low-resolution monitoring recordings.
Everyone can play pixelated at this address .
I wanted to test this tool myself in practice. Unfortunately, it did not work, but the results of their attempts are shared by many people on Twitter. While their posts are real, the results can be very funny. What I like most is the fun of creating realistic faces of the characters of old computer games. Who remembers these gentlemen?
Some say that StyleGAN was taught based on the white face pattern and did not do well with people of other races. For example, pixelated photos of Barack Obama, pixelated as a completely different man with a white face.
However, there are other examples proving that the application can do it much, much better. One of the photos is even hit.
If you are interested in such experiments, then it's worth looking at the creator's profile
Malimonov regularly publishes on his Twitter profile a lot of interesting materials using the growing power of artificial intelligence . Here are some examples:
Artificial intelligence turns pixelose into realistic portrait photos
Dear guys! Our prices may be slightly higher than other platforms, but we guarantee that your article is 100% unique and personalized. If other expert services charge very low fees, this means that creators are not being paid enough.
ReplyDelete