#mlgeneratedimages

We officially can no longer trust anything we see on the internet. AI is starting to distort reality — all with the help of machine learning. Researchers at NVIDIA have harnessed the power of a GAN — a class of neural network — to generate some extremely realistic faces. The results are more impressive than anything we’ve seen before. These AI generated images are viral now.

Generative Adversarial Network (GAN)

A GAN can iteratively generate images based on genuine photos it learns from. Then it evaluates the new images against the original. In this instance, the researchers taught a GAN a number of “styles”. For example, faces modeled after subjects who were old, young, wearing glasses, or had different hairstyles. The results are spectacular. Even small seemingly random details like freckles, skin pores or stubble have convincing distribution in the images the project generated. Behind the new feature is a technique NVIDIA calls “style-mixing.”   “To further encourage the styles to localize, we employ mixing regularization, where a given percentage of images are generated using two random latent codes instead of one during training. When generating such an image, we simply switch from one latent code to another — an operation we refer to as style mixing — at a randomly selected point in the synthesis network.” - From the paper.  

Stochastic Variation

Stochastic variation is another key property allowing GANs to realize the randomization of detailed facial features, such as the placement of facial hair, stubble density, freckles, pores, etc. The paper proposes adding per-pixel noise after each convolution layer. The added noise does not affect the overall composition or the high-level attributes of images. And the changing noise in different layers produces matching stochastic variation results. To quantify interpolation quality and disentanglement, the paper proposes two new, automated methods — perceptual path length and linear separability — that are applicable to any generator architecture. Researchers saw impressive results using the new generator to forge images of bedrooms, cars, and cats. It was developed with the Large-scale Scene Understanding (LSUN) dataset. Alongside today’s paper, NVIDIA has also released a huge new dataset of human faces. Flickr- Faces-HQ (FFHQ) contains 70,000 high-quality images at 1024 resolution. The dataset will soon be available to the public. Although Nvidia’s results may be the best so far, there are still tiny discrepancies if you stare long enough.

Recognizing The Images

Kyle McDonald, an artist working with code. He published a list of things to look out for when trying to recognize an image generated by an AI system. “At low resolutions, almost all the images in the paper are indistinguishable from photographs. There are only a few artifacts that stand out to me that I will try to address,” he said in a blog post. McDonald points out what he calls “the missing earring” problem. Often in the images, there is small circular glitch below the ears. This comes from the GAN attempting to add in earrings previously seen in photos. There are other slight oddities too. For example, like the asymmetry of facial features, lack of detailing in teeth, strange hair strands, and patterns in clothing.

The Future