High-quality facial attributes editing results from InterFaceGAN.
GAN Inversion
GAN inversion aims to invert a given image back into the latent space of a pre-trained GAN model, for the image to be faithfully reconstructed from the inverted code by the generator. This allows us to manipulate the latent vectors to control facial attributes like age, smile, etc by traversing along the corresponding dimensions. The following figure explains the high-level concept of GAN inversion and its use for face manipulation.
GAN Inversion Overview
GAN inversion for face manipulation:
We applied some of the state of art approaches in GAN inversion literature to invert an image to a latent code that supports high-fidelity controllable edits and realistic image generation. We ran the following approaches that gave us the following results:
- HyperStyle:
2. Pivotal Tuning Inversion:
3. StyleCLIP: