AI generation of fake celeb images

The dataset of 200,000 celeb photos with the face nicely centered at a known location is a nontrivial part of making the exercise feasible.

I trained on windows with a 6Gb 1060, and went off script from the DCGAN paper, by using upscaling rather than transpose convolutions. Once all the fiddling details are set correctly, the results are quite amazing. It didn’t even require a complete single pass over that dataset.

Generative models like GANs are fascinating, but very temperamental to train. Some of the findings in this paper mirror my own observations - increasing the complexity of the GAN adds a lot of instability. My solution was to keep things as simple as possible. I spent a lot of effort trying to increase the size of the network to get better results, but in the end my smallest implementation worked the best.

This bit is interesting: "Without progressive growing, all layers of the generator and discriminator are tasked with simultaneously finding succinct intermediate representations for both the large-scale variation and the small-scale detail. With progressive growing, however, the existing low-resolution layers are likely to have already converged early on, so the networks are only tasked with refining the representations by increasingly smaller-scale effects as new layers are introduce"

[0] https://www.youtube.com/watch?v=G06dEcZ-QTg&feature=youtu.be

https://www.nvidia.com/en-us/data-center/dgx-1/#order-now-dg...

The NVIDIA DGX is available for purchase in select countries and is priced at:

    DGX with P100 at $129,000*
    DGX with V100 at $149,000*
    DGX support plan is required and must be purchased separately.

The female's left and right eyes are different shapes, as are her eyebrows. And the male's ears appear to be in different places on the left and right side of his head. His eyes are creepily different too.

Nvidia article: https://youtu.be/G06dEcZ-QTg

Cry: https://youtu.be/KxtPRF6NG7I