Can a model be trained to take a specific face image and turn it into 3D selfie? Or all these images represent randomly generated people? Can it work not with faces only, but with whole human body?
I think you could keep the weights frozen and optimize over the latent space to obtain the latent that approximates your face. This is known as inversion / GAN inversion. You can read more about it here.
Can a model be trained to take a specific face image and turn it into 3D selfie? Or all these images represent randomly generated people? Can it work not with faces only, but with whole human body?