Admit it; we’ve all wondered what it would feel like to live in a world of our favorite anime. What would it feel like to eat ramen with Naruto? Or, how would the world of Fairytail look? Well, unfortunately, we’re not there yet; but, we are now one step closer to this reality thanks to NCSoft’s AI-powered anime converter.
NCSoft, the company behind the award-winning game Guild Wars 2, has funded research to create an application that turns anyone’s everyday picture to an anime version of themselves. Using the machine learning algorithm called Generative Adversarial Networks (GANs), the system collates data from thousands of inputs like human and anime character faces.
Junho Kim, one of the lead researchers, credits a paper called ‘class activation mapping’ (CAM) as one of the inspirations for the software. CAM organizes and groups points of interest from images that the user feeds the system. The NCSoft team of researchers applied this same concept by having their system use “attention maps.”
This function will allow the system to identify characteristics that will differentiate one image from another. The application also a normalization algorithm called Adaptive Layer-Instance Normalization (AdaLIN) to produce convincing and amazing anime portraits of people.
Besides human subjects, the application can also transform other types of images. For example, it can convert an animal such as a dog to a completely unrelated animal such as a cat. However, only really powerful computers can run the system as of the moment. So, don’t worry if your computer crashes if you attempt to use it. Nevertheless, it won’t be too long before everyone is doing such transformations with just a click of a button.
NCSoft’s Hyeonwoo Kang, Junho Kim, Kwanghee Lee, and Minjae Kim are credited for the creation of the software. The application is available in Github for software enthusiasts and researchers.
