# TU Wien Rendering #26 - Low Discrepancy Sequences ## Метаданные - **Канал:** Two Minute Papers - **YouTube:** https://www.youtube.com/watch?v=1ziudxJT884 - **Дата:** 15.05.2015 - **Длительность:** 4:55 - **Просмотры:** 5,625 ## Описание In this segment we explore a subset of Quasi-Monte Carlo methods called low discrepancy series. Examples of this are the Halto and Van der Corput series. These are deterministically generated sample sequences that stratify well even in high dimensional Euclidean spaces. Surprisingly, randomly generated samples don't have this desirable property! About the course: This course aims to give an overview of basic and state-of-the-art methods of rendering. Offline methods such as ray and path tracing, photon mapping and many other algorithms are introduced and various refinement are explained. The basics of the involved physics, such as geometric optics, surface and media interaction with light and camera models are outlined. The apparatus of Monte Carlo methods is introduced which is heavily used in several algorithms and its refinement in the form of stratified sampling and the Metropolis-Hastings method is explained. At the end of the course students should be familiar with common techniques in rendering and find their way around the current state-of-the-art of the field. Furthermore the exercises should deepen the attendees' understanding of the basic principles of light transport and enable them to write a simple rendering program themselves. These videos are the recordings of the lectures of 2015 at the Teschnische Universität Wien by Károly Zsolnai and Thomas Auzinger Course website and slides → http://www.cg.tuwien.ac.at/courses/Rendering/ Subscribe → http://www.youtube.com/subscription_center?add_user=keeroyz Web → https://cg.tuwien.ac.at/~zsolnai/ Twitter → https://twitter.com/karoly_zsolnai ## Содержание ### [0:00](https://www.youtube.com/watch?v=1ziudxJT884) Segment 1 (00:00 - 04:00) it's not about discrepancy series what we have been talking about so far is that about something this means if I have the random number generator in general examples and this is the sample that I'm going to use and many additions were thinking that we could perhaps too much better than that because what I would be looking for is Hemisphere and I'll be shooting samples on the surface on this Hemisphere and create unit vectors on the surface of this Hemisphere and I can do this deterministically what if I have an algorithm that doesn't generate random numbers but it will make sure that this time is the a if I have 100 samples the samples are well distributed on this hemisphere and if you do this you may get much better convergence and much better looking result so below you can see a random number generator generating samples in candy and after the whole tone sequence which is called enrollee frequency series what this means is it's not completely randomly but it tries to fill the space reasonably but this isn't map video if you read after it what you would think is that it would be very simple to get a grid and just put points in the mid point and then you would have samples are really wellness committee and this you can do it truly in one line stakes of this works there but there are mathematical rules and that this is absolutely certain in higher dimension so if you're higher dimensional spaces then this is a 12 stratified so the most commonly used sequences are the outer series so whole series of water quality series we can do this little discrepancy sampling in many different ways and this Matta gives you an even distribution of the noise because you are sampling these hemispheres reasonably stratified with so it cannot really be that one side of the hemisphere sampled almost exhaustively and the other one is completely neglected so you would get images with a noise distribution that's better for you so that's a plus but what is even more important is that it is deterministic so if you render it an animation imagine completely rendom something we have frame number one you distribute your suppose that comes very manually and then you distribute your samples in a completely different way so the noise would look like this of anyone and death completely different so until you have converged perfectly you will have these issues that before tako flickering with a coherence issues because the noise looks like this on frame or looks like that of a loop and if you pay twenty five of these frames every second or maybe even more then you will add with really this problem in crater you are computing everything and the football series and all discrepancy series help you with them so they like to use the same industry because of this reason because even subsequent frames will compute the very same advantages okay disadvantages well disadvantages are also huge it's often map reveal if you take a look at this image these walls are not textured these are one color dream this is one color green if used well and this is a one power you shouldn't feel a bit like this at all this is a image and I have implemented the amount of support and the problem that I encountered was for eventing conventions and this is a serious problem that you can encounter I will not go into the details but you just mess up with one small detail and you can get an image like that well this is actually the delightful way of failing I don't know about you but most of my primary areas are necessarily supportive like this they tend to look like this so I usually segmentation and stuff like that so it's if you make a mistake in mobile illumination wondering even your errors but you better than in other fields --- *Источник: https://ekstraktznaniy.ru/video/14986*