We have an algorithm nowadays that help us to draw a picture of a 3D object. This algorithm takes an image of a 3D model and gives a step-by-step breakdown to draw the image of that 3D object. This algorithm uses graph theory to break down an image into smaller pieces.
But if you go in the opposite direction, we also
have an algorithm that gives a photo-realistic image from a rough drawing as an
input. That looks science fiction but scientists at NVIDIA are already using it
for 1.5 years.
In this, we can pick what we want to draw from the options given below on the left side and get an output on the right side as a hopefully photo-realistic landscape image.
But, it has a limitation that it can't synthesize human faces. But with the technique that we are going to talk about today, this problem is being solved. This is called
DeepFaceDrawing: Deep Generation of Face Images from Sketches
Creating realistic human face images from scratch benefits various applications including criminal investigation, character design, educational training, etc. Due to their simplicity, conciseness, and ease of use, sketches are often used to depict desired faces. The recently proposed deep learning-based image-to-image translation techniques allow automatic generation of photo images from sketches for various object categories including human faces, and lead to impressive results.
In this, you just have to draw a freehand sketch as a guide image and you will get a nearly photo-realistic human face that matches your sketch. Interestingly, it also gives you something as a starting point before you draw the hair itself. But, if you choose to give your desired hairstyle, it will follow your drawing. Isn't it amazing?
That's not it, it also gives you many more amazing features that help you to adjust different facial features and your input can be mapped to many possible human faces.
It has another amazing feature, where, even you don't need to draw anything. You just have to take few aspects of human faces that you like to combine. And you get your desired human face. This is called
Face Copy-Paste
But this AI got few bugs, and that is, its inability to match eye color and asymmetry in the output face as you can see in the picture. But it’s a just matter of time to get that problem to solve.
I am sure digital artists are gonna take this technique in near future and it will help artists to spend more time bringing their artistic vision to life and spend less time on the execution.
