Technology makes it possible to transform selfies into anime characters, cats in dogs, horses in zebras or landscapes in impressionist-style paintings
Let’s face it, we’re obsessed with our pets! From our ridiculous nicknames, to dressing them up for Halloween, we can’t get enough of our dogs and cats! If you’re anything like us, you’ve dreamed of cartoonizing your dog or cat into a fun caricature or portrait to capture their unique personality. Here’s a handy DIY guide to cartoonize your adorable pet cat or dog! You may also want to check out our guide to turning your pet into a bitmoji as well!
After cartoonizing your pet you can also create a custom Petventures storybook featuring your pet or custom pet pop art blankets, canvas, socks, and phone cases.
The virality of photographs where someone’s face is changed in some way has prompted the emergence of new applications that seek to expand the terrain already explored and transform the face of users into other types of images.
Among the best known is FaceApp, capable of changing the face of a person to make them smile, look younger, older or change sex. It was developed by the Russian company Wireless Lab and launched in 2017.
It may not be an App to use every day, but it is not bad to create different images and with funny results. It is compatible from Android 2.3, although it weighs 38 megabytes, so some short memory devices will suffer to install it. And recently many social network users have shared images of themselves, their families or their partners with a particular edition: portrayed as cartoons that highlight the faces of the protagonists.
That happened after the arrival of MomentCam, an application to animate the photographs through various characters and situations. Although the app was launched last November by creatives in China and later popularized in other countries – also delivering an English version – in Colombia it has had a boom only until these last weeks.
Another one that has gained fame in recent weeks is Al Portraits that turns photos into pictures of different artistic styles and levels of abstraction. It is developed by the IBM Watson Artificial Intelligence (AI) Laboratory of the Massachusetts Institute of Technology (MIT).
A new project with no name comes to this already explored terrain, but promising by the developers behind it, the NCSoft video game company, responsible for titles such as “Lineage” and the “Guild Wars” saga.
NCSoft developed an artificial intelligence that transforms the face of people into anime characters and the face of pets changes it to that of another animal.
The tool is not the easiest to use, since the software is only compatible for PCs enabled to run open source models, in addition the user will have to train it on their own. But with the project so advanced, it is likely that NCSoft itself or other independent developers will simplify it for immediate use.
Turning human faces into animated is simple, but to achieve this, different algorithms that work together are required in order to obtain similar images. The software with which it was made is an open source based on antagonistic generative networks (better known by its acronym in English GANs), that is, they are two networks where one generates samples of what you want to create and the other discriminates the attempts of the first until they give the desired result.
The rise of GANS dates back five years, but one of its latest applications in technology is to generate hyperrealistic human faces from thousands of photos.
Those of NCSoft used the GANs so that the algorithm would learn from a large number of images of animated faces and from them create an anime drawing similar to the user’s selfie.
The researchers responsible for the tool are Junho Kim, Minjae Kim, Hyeonwoo Kang and Kwanghee Lee, who are also responsible for the development of “Guild War 2”.
A attention mapping code was added to the GAN model, where an image is divided into layers to highlight the most important elements, a procedure known as class activation mapping or CAM.
CAMs allow you to train a neural network with two images to detect the similarities, classify them and use them to create a new one. If the photo of a dog and a cat is loaded, for example, it will identify that both have similar hair, ears and eyes, it will take those characteristics and gather them together to make it a single portrait.
The developed ones incorporated the attention maps to the GAN model so that the algorithm focused on important regions of the images and discarded the less relevant ones. Finally, they added a normalization code, known as adaptive layer-instance normalization (AdaLIN) for reliable and convincing results.
All this technology makes it possible to transform selfies into anime characters, cats into dogs, horses by zebras or landscapes in impressionist-style paintings.
The tool is available on GitHub and is free. The details of the research were published on the online scientific article platform, ArXiv, so anyone interested can read it. Download MomentCam App Store / Google Play