mediapipe VS stylegan2-pytorch

We've trained ControlNet on a subset of the LAION-Face dataset using modified output from MediaPipe's face mesh annotator to provide a new level of control when generating images of faces.

Medipaipe library link: https://mediapipe.dev/

If you're looking for the differences in terms of how inference is done, I recommend you take a look at MediaPipe's source code. MediaPipe doesn't use raw code, but uses a "graph" instead (eg. pose_landmark_cpu.pbtxt), which can be visualised using MediaPipe Viz. I also used axinc-ai/ailia-models as the base (preprocessing, inference, postprocessing, etc...) which I further built upon (keypoint refinement, roi from keypoints, filtering / smoothing, etc...)

I thought of doing that but unfortunately medipipe requires a RGB input and performs better with it more on that here

You might want to check out https://github.com/open-mmlab/mmhuman3d and https://github.com/google/mediapipe

GANs on the other hand uses a generator and discriminator model, where the generator learns to generate images that are "good enough" to fool the discriminator. It may make it easier to make discriminators that are better at picking out visual artefacts that stand out for humans, (it was easy to pick out "fake humans" from the early versions of thispersondoesnotexist.com by looking at their eyes and teeth, they tend to be pointing in different directions, or their teeth are too irregular before the discriminator got better at picking those out) but they can suffer from mode collapse, and generate images with very little variety.

Generative networks like StyleGAN

It's interesting that AI generated pictures of people can get most things right but seemingly always struggles with teeth & fingers. If you check out thispersondoesnotexist.com the teeth are almost always goofy looking...

Teeth as well, check out thispersondoesnotexist.com you'll come across a lot of convincing looking people with odd teeth

https://thispersondoesnotexist.com/ has been around for 4 years now...

I remember there was https://thispersondoesnotexist.com/ Did it also work on StableDiffusion beneath?

Another great place to see AI generated faces: https://thispersondoesnotexist.com/

Check out https://thispersondoesnotexist.com/