Deep learning algorithms are at the forefront of machine learning and AI, and the applications are endless. A team from the Chinese Academy of Sciences in Beijing has developed DeepFaceDrawing, which uses AI to collect as much possible information from a simple hand drawn sketch as possible and produces a photo-realistic picture of a person’s face.
Read: Competition: LG Tone Ultra HBS-810 Headset Giveaway
According to the team the AI helps “users with little training in drawing to produce high-quality images from rough or even incomplete freehand sketches.” While similar tools have been created in the past, this is the most advanced of its kind. It requires less detail in sketches than its predecessors to create a relatively accurate representation of the face. Sure, there may be some incorrect approximations (called false positives) but it is the best of its kind.
So how would such an AI work, you ask? Through deep learning it is able to work out how different components of the face generally relate to each other. This means you don’t need an accurate representation in the sketch of an eye, for example, because it has learned how an eye is most likely to appear taking the mouth, nose, eyebrows into consideration.
It then chooses a look that has the highest probability given the mix of facial components, as it scours through an extensive list of reference faces. So the final product will never be the face of an actual person in the database, but rather a conglomeration of the various facial components most likely to occur together given what can be seen in the sketch.
According to the research, “recent deep image-to-image translation techniques allow fast generation of face images from freehand sketches. However, existing solutions tend to overfit to sketches, thus requiring professional sketches or even edge maps as input. To address this issue, our key idea is to implicitly model the shape space of plausible face images and synthesize a face image in this space to approximate an input sketch. Our method essentially uses input sketches as soft constraints and is thus able to produce high-quality face images even from rough and/or incomplete sketches.”
A pertinent question that has been asked, especially with everything that is going on around the world with the Black Lives Matter movement, is how it will handle race. We’ve seen in the past that these and traditional facial recognition technologies tend to identify non-white faces much less accurately. The majority of the dataset that was used to train this AI was reportedly from white and South American faces, so we can expect similar issues here.
