Dr. Zsolnai demos a new AI image editor that lets you adjust faces, cars, and other objects with text commands or sliders. He transforms himself into Obi-Wan Kenobi and shows how the technique works on multiple object types.
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today's paper is about creating synthetic human faces. And not only that, but it can also make me look like Obi Wan Kenobi. You will see the rest of this footage in a few minutes. Now, of course, this is not the first paper to generate artificial human faces. For instance, in December 2019, a technique by the name Stargan 2 was published. This is a neural network based learning algorithm that is capable of synthesizing these eye poppingly detailed images of human beings that don't even exist. This work answered some questions and as any good paper raised many more good ones. For instance, generating images of virtual humans is fine, but what if the results are not exactly what we are looking for? Can we have some artistic control over the outputs? How do we even tell the AI what we are looking for? Well, we are in luck because Stalgen 2 offers somewhat rudimentary control over the outputs where we can give it input images of two people and fuse them together. Now that is absolutely amazing. But I wonder if we can ask for a little more. Can we get even more granular control over these images? What if we could just type in what we are looking for and somehow the AI would understand and execute our wishes? Is that possible? Or is that science fiction? Well, hold on to your papers and let's see. This new technique works as we type what aspect of the input image we wish to change and what the change should be. Wow, really cool. And we can even play with these sliders to adjust the magnitude of the changes as well. This means that we can give someone a new hairstyle, add or remove makeup, or give them some wrinkles for good measure. Now the original Stalgen 2 method worked on not only humans, but on a multitude of different classes too. And the new technique also inherits this property. Look, we can even design new car shapes, make them a little sportier, or make our adorable cat even cuter. For some definition of cuter, of course, we can even make their hair longer. Or change their colors. And the results are of super high quality. Absolutely stunning. While we are enjoying some more results here, make sure to have a look at the paper in the video description. And if you do, you will find that we really just scratched the surface here. For instance, it can even add clouds to the background of an image, or redesign the architecture of buildings and much, much more. There are also comparisons against previous methods in there, showcasing the improvements of the new method. And now let's experiment a little on me. Look, this is me Here after I got locked up for dropping my papers and I spent so long in there that I grew a beard. Or I mean a previous learning based AI called StyleFlow gave me one. And since dropping your papers is a serious crime. The sentence is long. Quite long. Ouch. I hereby promise to never drop my papers ever again. So now let's try to move forward with this image and give it to this new algorithm for some additional work. This is the original, and by original I mean the image with the added algorithmic beard from a previous AI. And this is the embedded version of the image. This image looks a little different. Why is that? It is because Stalgean 2 runs an embedding operation on the photo before starting its work. This is its own internal understanding of my image, if you will. This is great information and is something that we can only experience if we have hands on experience with the algorithm. And now let's use this new technique to apply some more magic to this image. This is where the goodness happens. And oh my, it does not disappoint. You see, it can gradually transform me into Obi Wan Kenobi. An elegant algorithm for a more civilized age. But that's not all. It can also create a ginger Caroj hippie, Caroi Caroj who found a shiny new paper on fluid simulations and Karoj who read sat paper outside for quite a while and was perhaps disappointed. And now hold on to your papers and please welcome Dr. Carolina Zsolnai. And one more. I apologize in advance. Rockstar Karoj with a mohawk. How cool is that? I would like to send a huge thank you to the authors for taking their time out of their workday kids to create these images only for us. You can really only see these here on two minute papers. And as you see, the pace of progress in machine learning research is absolutely stunning. And with this, the limit of our artistic workflow is not going to be our mechanical skills, but only our imagination. What a time to be alive. This video has been supported by Weights and Biases. Check out their recent offering Fully Connected, with a place where they bring machine learning practitioners together to share and discuss their ideas, learn from industry leaders and even collaborate on projects together. You see, I get messages from you fellow scholars telling me that you have been inspired by the series but don't really know where to start. And here it is. Fully Connected is a great way to learn about the fundamentals, how to reproduce experiments, get your papers accepted to a conference and more. Make sure to visit them through WNB me papers or just click the link in the video description. Our thanks to WeightsAndBiases for their long standing support and for helping us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
