# Opening The First AI Hair Salon! 💇

## Метаданные

- **Канал:** Two Minute Papers
- **YouTube:** https://www.youtube.com/watch?v=0ISa3uubuac
- **Дата:** 26.01.2022
- **Длительность:** 5:14
- **Просмотры:** 44,867
- **Источник:** https://ekstraktznaniy.ru/video/13684

## Описание

❤️ Check out Weights & Biases and say hi in their community forum here: https://wandb.me/paperforum

📝 The paper "SketchHairSalon: Deep Sketch-based Hair Image Synthesis" is available here:
https://chufengxiao.github.io/SketchHairSalon/

🙏 We would like to thank our generous Patreon supporters who make Two Minute Papers possible:
Aleksandr Mashrabov, Alex Balfanz, Alex Haro, Andrew Melnychuk, Angelos Evripiotis, Benji Rabhan, Bryan Learn, Christian Ahlin, Eric Martel, Gordon Child, Ivo Galic, Jace O'Brien, Javier Bustamante, John Le, Jonas, Jonathan, Kenneth Davis, Klaus Busse, Lorin Atzberger, Lukas Biewald, Matthew Allen Fisher, Mark Oates, Michael Albrecht, Michael Tedder, Nikhil Velpanur, Owen Campbell-Moore, Owen Skarpness, Peter Edwards, Rajarshi Nigam, Ramsey Elbasheer, Steef, Taras Bobrovytsky, Thomas Krcmar, Timothy Sum Hon Mun, Torsten Reil, Tybie Fitzhugh, Ueli Gallizzi.
If you wish to appear here or pick up other perks, click here: https://www.patreon.com/TwoMinutePapers

T

## Транскрипт

### Segment 1 (00:00 - 05:00) []

Dear Fellow Scholars, this is Two Minute Papers with Dr. Károly Zsolnai-Fehér. Today we are going to open an AI-powered hair salon. How cool is that. But, how? Well, we recently looked at this related technique from NVIDIA that is able to make our rough drawings come to life as beautiful photorealistic images. In goes a rough drawing, and out comes an image of this quality. And it processes changes to the drawings with the speed of thought. You can even request a bunch of variations on the same theme, and get them right away. So here is a crazy idea - how about doing this with hair styles? Just draw them, and the AI puts it onto the model and makes it look photorealistic. Well, that sounds like science fiction, so I will believe it when I see it. And, this new method claims to be able to do exactly that, so, let’s see the pros first. This will go in, and this will come out. How is that even possible? Well, all we need to do is once again, produce a crude drawing that is just enough to convey our intentions. We can even control the color and the length of the hair locks, or, even request a braid, and all of these work pretty well. Another great aspect of this technique is that it works very rapidly, so we can easily iterate over these hair styles, and if we don’t feel like our original idea will be the one, we can refine it for as long as we please. The AI too works with the speed of thought. One of the best aspects of this work. You can see an example here. This also does not look right…yet, but this quick iteration also means that we can also lengthen those braids in just a moment. And, yes, better! So, how does it perform compared to previous techniques. Well, let’s have a look together! This is the input hairstyle, and a matte telling the AI where we seek to place the hair, and, let’s see the previous methods. The first two techniques are very rough, this followup work is significantly better than the previous ones, but it is still well behind the true photograph for reference. This is much more realistic. So, hold on to your papers, and let’s see the new technique. Now you’re probably thinking Károly…what’s going on? Why is the video not changing? Is this an error? Well, I have to tell you something. It’s not an error. What you are looking at now is not the reference photo. No-no. This is the result of the new technique. The true reference photo is this one. That is a wonderful result. The new method understands not only the geometry of the hair better, but also how the hair should play with the illumination of the scene too. I am a light transport researcher by trade, and this makes me very, very happy. So much so, that in some ways, this technique looks more realistic than the true image. That is super cool. What a time to be alive! Now, not even this technique is perfect. Even though the new method understands geometry better than its predecessors, we have to be careful with our edits, because geometry problems still emerge. Look here. Also, the other issue is that the resolution of the generated hair should match underlying image. I feel that these is usually a bit more pixelated. Now note that these kinds of issues are what typically get improved from one paper to the next, so, you know the deal, a couple more papers down the line, and I am sure, these edits will become nearly indistinguishable from reality. As you see, there are still issues, the results are often not even close to perfect, but, the pace of progress in AI research is nothing short of amazing. With this, we can kind of open an AI-powered hair salon, and some cases will work, but soon, it might be that nearly all of our drawing will result in photorealistic results that are indistinguishable from reality.

### Segment 2 (05:00 - 05:00) [5:00]

Thanks for watching and for your generous support, and I'll see you next time!