An image of @BarackObama getting upsampled into a white guy is floating around because it illustrates racial bias in #MachineLearning. Just in case you think it isn't real, it is, I got the code working locally. Here is me, and here is @AOC.

Here is my wife @shan_ness

This is @LucyLiu

I am pointing out the limits and social implications of deep generative models. I'm not deriding the academic contributions of these authors. Deep gen modeling lit is a bit hypish, showing their limitations helps us understand how they could be made useful https://arxiv.org/abs/2003.03808 

Hey @samcharrington this is the best one, I think you should have it framed.

By popular demand, here is an example of a white person.
Perhaps we can power through the mismatch with more compute.

But more to the point, Marylin didn't mysteriously turn into a black or latinx person.

OK ok one more. @Abebab wins the prize for most hilarious transformations.

Researchers in deep generative modeling are obsessed with generating photo-realistic images from more abstract/low-information representations (down-sampled, cartoons, sketches, etc.)

This is problem.

At best, the frivolity obscures the technical contribution. At worst, it encourages dangerous applications.

For example, this paper uses StyleGAN to "beautify" faces. Why spend effort on this problem? I dare you to try it on a dark skinned person. https://deepai.org/publication/gan-based-facial-attractiveness-enhancement

My call to the community is to think more deeply about modeling and disentangling abstraction and meaning in an image. That gets us closer to more interesting use cases.

I'm reminded of this figure by @scottmccloud