A new AI method generates realistic hairstyles on virtual humans with separate control over shape and color. It smoothly morphs between styles while keeping facial details sharp, fixing the artifacts that plagued earlier techniques. The researchers provided exclusive examples showing the quality jump.
Dear fellow scholars, this is two minute papers with Dr. Zsolnai, AI research is improving at such a rapid pace that today we can not only create virtual humans, but even exert some artistic control over how these virtual humans should look. But today we can do something even cooler. What can be even cooler than that? Well, have you ever seen someone with a really cool haircut and wondered what you would look like with it? Would it be ridiculous or would you be able to pull it off? Well, hold on to your papers because you don't have to wonder anymore. With this new AI, you can visualize that yourself. Hmm. Ok, but wait a second. Previous methods were already able to do that. So. So what is new here? Well, yes, but look at the quality of the results. In some cases it is quite easy to find out that these are synthetic images, while for other cases, so many of the fine details of the face get lost in the process that the end result is not that convincing. So is that it is the hairstyle synthesis Dream dead. Well, if you have been holding on to your papers, now squeeze that paper and have a look at this new technique. And wow. My goodness, just look at that. This is a huge improvement. Just one more paper down the line. And note that these previous methods were not some ancient techniques. No, no. These are from just one and two years ago. So much improvement in so little time. I love it. Now let's see what more it can do. Look, we can choose the hairstyle we are looking for. And yes, it doesn't just give us one image to work with. No. It gradually morphs our hair into the desired target shape. That is fantastic. Because even if we don't like the final hairstyle, one of those intermediate images may give us just what we are looking for. Loving it. And it has one more fantastic property. Look. Most of the details of the face remain in the final results. Whew. We don't even have to morph into a different person to be able to pull this off. Now, interestingly, look, the eye color may change, but the rest seems to be very close to the original. Not perfect, but very close. And did you notice? Yes, it gets better. Way better. We can even choose the structure. This would be the hairstyle, which can be one image. And the appearance, this is more like the hair color. So does that mean that. Yes, we can even choose them separately. That is, take two hairstyles and fuse them together. Now have a look at this too. This one is an exclusive example. To the best of my knowledge, you can only see this here on 2 Minute Papers. A huge thank you to the authors for creating these just for us. And I must say that we could stop on any of these images and I am not sure if I would be able to tell that it is synthetic. I may have a hunch due to how flamboyant some of these hairstyles are, but not due to quality concerns, that's for sure. Even if we subject these images to closer inspection, the visual artifacts of the previous techniques around the boundaries of the hair seem to be completely gone and I wonder what this will be able to do just a couple more papers down the line. What do you think? I'd love to know. Let me know in the comments below. What you see here is a report of this exact paper we have talked about which was made by Weights and Biases. I put a link to it in the description. Make sure to have a look. I think it helps you understand this paper better. Weights and Biases provides tools to track your experiments in your deep learning projects. Using their system you can create beautiful reports like this one to explain your findings to your colleagues better. It is used by many prestigious labs including OpenAI, Toyota Research, GitHub and more. And the best part is that Weights and Biases is free for all individuals, academics and open source projects. Make sure to Visit them through wnb.com papers or just click the link in the video description and you can get a free demo today. Our thanks to Weights and Biases 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.
