Need a tool that can create a realistic rendition of a human’s face in cartoon form? Microsoft AI can get the job done for you with ease. Let’s take an in-depth look at what it’s capable of.
Becoming a great cartoonist is a skill that takes years to master. However, Microsoft engineers have automated some of the process via generative adversarial networks. Although the end result is not exactly the real thing, it’s a realistic-enough looking fake that’s quite impressive and looks like something a real artist would produce.
When generating the image, there are a couple of things you can tweak. For instance, the user has a choice in determining how exaggerated the face will end up looking. If you’re after a particular style, you can also supply an example caricature the AI will attempt to mimic.
To test how distinguishable the results are, Microsoft conducted a test in which participants were asked to identify the correct one. They were also asked how faithful to the original the end results looked in comparison to hand-drawn caricatures, and the AI-generated results were favoured 22.95% of the time.
In conclusion, the AI output is sometimes completely indistinguishable from real hand-drawn images. However, the researchers noted that an ideal fooling rate would be around 50%.
If you’re interested in the project and its full scope of findings and results, head on over to the CariGAN project page to read all about it.