Computer scientists at Microsoft and Hong Kong City University have developed two separate AI systems, forming a type of artificial neural network called Generative Adversarial Network (GAN) capable of caricatures. .
From uploaded portrait photos, one of the two GAN networks will analyze and magnify certain facial features. The rest of the GAN will add some stylistic strokes often found in caricatures, making portrait photography an interesting cartoon.
The interesting thing about these algorithms is that they can work in reverse, converting caricatures into lifelike images like photos.
However, this system still has some limitations. Compared to other facial features, the degree of geometric magnification is more pronounced in the overall face shape. The algorithm can only read 33 out of 63 landmarks located on the face block so it is not possible to apply some level of detailed geometric magnification on the ears, hair, wrinkles … The researchers said, they can address this limit by adding more landmarks on the face.
This AI system can even create caricatures in the form of video frame by frame.
- Discover the portal of artificial intelligence experience VTCC.Ai of the Cyber Center