|Facebook had become significant in one's lifestyle / Photo by: tongcom photographer via Shutterstock|
For several years, technological innovations made it possible to edit a photo to look better or make something fake become real. However, among all challenges in photo editing, the replacement of closed eyes with open ones in snapshots proved to be difficult, until Facebook cracked it. A research by the company that applied artificial intelligence managed to open closed eyes in photos in a pretty convincing manner.
The AI-based research utilizes a class of AI algorithms called Generative Adversarial Networks. GANs is powered by a system with two neural networks that contend with each other in a zero-sum game framework. In Facebook, GANs simply tries to fool itself into thinking that everything it creates is real.
The system works in two ways in order to achieve this behavior. First, one part of the AI learns to recognize a specific matter or topic like faces. Second, the other part of the AI generates images continuously based on the feedback obtained from facial recognition.
The repeated facial recognition and image generation of the two-part system of GANs fooled the AI and enabled potential realistic results. If applied to photos, the repetitive process replaces the closed eyes with open ones.
Unfortunately, standard machines running RANs do not understand intuitive factors like skin color when a person opens or closes their eyes, which could result in flawed editing. To circumvent this problem, Facebook introduced Exemplar GANs, a conditional GANs, to train the AI to understand the aspects of a person’s eyes, such as color, shape, and others. The result of the ExGANs for eye replacement is exceptionally realistic that even people mistook the fake eyes as real ones.
“We show that ExGANs can produce photo-realistic personalized in-painting results that are both perceptually and semantically plausible by applying them to the task of closed-to-open eye in-painting in natural pictures,” Facebook researchers noted.