Adobe’s New AI: Next Level Cat Videos! 🐈

Dear Fellow Scholars, this is Two Minute  Papers with Dr. Károly Zsolnai-Fehér. As you see here, I love Two Minute Papers, but  fortunately I am not the only one, Elon Musk   loves Two Minute Papers too. What’s more, even  cats love it. Everybody loves Two Minute Papers.    So, what was that? Well, these are all synthetic  videos that were made by this new, magical AI. So, here, we are adding synthetic data to an  already existing photorealistic video, and the   AI is meant to be able to understand the movement,  facial changes, geometry changes, and track them   correctly and move the moustache, the tattoos, or  anything else properly. That is a huge challenge. How is that even possible? Many AI-based  techniques from just a year or two ago   were not even close to being able to pull this  off. However, now, hold on to your papers,   and have a look at this new technique, now that  is a great difference. Look at that! So cool! This is going to be really useful  for at the very least two things.    One, augmented reality applications. For instance, have a look at this cat. Well, the  bar on the internet for cat videos is quite high,   and this will not cut it. But  wait…yes! Now we’re talking! The other examples also showcase that the progress  in machine learning and AI research these days   is absolutely incredible. I would like to send a  big thank you to the authors for taking the time   off their busy day to create these results only  for us. That is a huge honor. Thank you so much! So far, this is fantastic news. Especially  given that many works are limited to one domain,   for instance, Microsoft’s human  generator technique is limited to people. However,   this works on people, dogs, cats, and even Teslas.   So good. I love it. What a time to be alive! But wait, I promised two applications,  not one. So, what is the other one?    Well, image editing. Image editing? Really? What  about that? That is not that new…but…aha! Not just   image editing, but mass scale image editing!   Everybody can get antlers, tattoos, stickers, you   name it. We just chuck in the dataset of images,  choose the change that we wish to make, and out   comes the entirety edited dataset automatically.   Now that is a fantastic value proposition. But, of course, you, the seasoned  Fellow Scholar immediately knows,   that surely, not even this technique is  perfect. So, where are the weak points? Look.    Oh yes, that. I see this issue for nearly  every technique that takes on this task.    It still has troubles with  weird angles and occlusions. But, we also have good news. If you can navigate  your way around an online Colab notebook,   you can try it too! You know what to  do. Yes, let the experiments begin! So, what would you use this for? Who would  rock the best moustache? And, what do you   expect to happen a couple more papers down the  line? Please let me know in the comments below. Thanks for watching and for your generous  support, and I'll see you next time!