Adobe's new technique adds detailed bumps and ridges to 3D models while cutting memory use from 2 GB down to 34 MB. The method works 50 times more efficiently than traditional approaches and lets artists render interactively. Dr. Zsolnai breaks down how it works.
February 12, 2022
34 megabytes
2 gigabytes
About 50 times less memory
20 views
Dear fellow scholars, this is Two Minute Papers with Dr. Zsolnai. Today we are going to create a virtual world with breathtakingly detailed pieces of geometry. These are just some of the results of the new technique and boy, are they amazing. I wonder how expensive this will get. We'll see about that in a moment. Now here we wish to create a virtual world and we want this world to be convincing. Therefore, we will need tons of high resolution geometry like the ones you see from these previous episodes. But all this detail in the geometry means tons of data that has to be stored somewhere. Traditional techniques allow us to crank up the detail in the geometry, but there is a price to be paid for it. And the price is that we need to throw more memory and computation at the problem. The more we do, the higher the resolution of the geometry that we get. However, here comes the problem. Ouch. Look at higher resolution levels. Yes, the price to be paid is getting a little steep. Hundreds of megabytes of memory is quite steep for just one object. But wait, it gets worse. Oh, come on. Gigabytes. That is a little too much. Imagine a scene with hundreds of these objects laying around. No graphics card has enough memory to do that. But this new technique helps us add small bumps and ridges to these objects much more efficiently than previous techniques. This opens up the possibility for creating breathtakingly detailed digital objects with where even the tiniest imperfections on our armor can be seen. Ok, that is wonderful. But let's compare how previous techniques can deal with this problem and see if this new one is any better. Here is a traditional technique at its best when using 2 gigabytes of memory. That is quite expensive. And here is the new technique. Look at that. This looks even better. Surely this means that it needs even more memory. How many gigabytes does this need? Or if not gigabytes, how many hundreds of megabytes? What do you think? Please let me know in the comment section below. I'll wait. Thank you. I am super excited to see your guesses. Now hold on to your papers because actually, it's not gigabytes. Not even hundreds of megabytes? No, no. It is 34 megabytes. That is a pittance for a piece of geometry of this quality. This is insanity. But wait, it gets better. We can even dramatically change the displacements on our models. Here is the workflow. We plug this geometry into a light simulation program and we can change the geometry, the lighting, material, properties, or even clone our object many, many times. And it still runs interactively. Now, what about the noise in these images. This is a light simulation technique called path tracing, and the noise that you see here slowly clears up over time as we simulate the path of many more millions of light rays. If we let it run for long enough, we end up with a nearly perfect piece of geometry. And I wonder how the traditional technique is able to perform if it is also given the same memory allowance. Well, let's see. Wow. There is no contest here. Even for other cases. It is not uncommon that the new technique can create an equivalent or or even better geometry quality, but use 50 times less memory to do that. So we might get more detailed virtual worlds for cheaper. Sign me up right now. What a time to be alive. And one more thing, when checking out the talk for the paper I saw this. Whoa. 20 views. 20 people have seen this talk now. I always say that views are of course not everything, but I am worried that if we don't talk about it here on two minute papers, almost no one will talk about it. And these works are so good. People have to know and if you agree, please spread the word on these papers and show them to people you know. This episode has been supported by Cohere AI. Cohere builds large language models and makes them available through an API so businesses can add advanced language understanding to their system or app quickly. With just one line of code, you can use your own data, whether it's text from customer service requests, legal contracts, or social media posts, to create your own custom models to understand text or even generate it. For instance, it can be used to automatically determine how whether your messages are about your business hours, returns, or shipping. Or it can be used to generate a list of possible sentences you can use for your product descriptions. Make sure to go to Cohere AI papers or click the link in the video description and give it a try today. It's super easy to use. Thanks for watching and for your generous support and I'll see you next time.
