Source: Deep Learning on Medium
Thank you for this detailed overview of StyleGAN2. Very inspiring read and crucial references!
I am also curious about the real use of projection, especially for forensics, since the images could be post-processed, or produced by complex StyleGAN2 implementations like Artbreeder (where you can apply various changes to various layers). I’ve tested one Artbreeder generated image (this application has already StyleGAN2 implemented for face generation) and even StyleGAN2 notebook implementations — the results are rather sobering, indeed.