A new technique renders cloth with hundreds of millions of individual fiber curves in real time. It generates fiber geometry directly on the GPU, handles self-shadowing through level-of-detail tricks, and supports multiple yarn models. The authors released cloth models for comparison.
Chapters
Dear fellow scholars, this is two minute papers with this piece of work shows us how to render a piece of cloth down to the level of fibers. This is a difficult problem because we need to be able to handle models that are built from potentially over a hundred million fiber curves. This technique supports a variety of goodies. One level of detail is possible. This means that the closer we get to the cloth, the more details appear and that it is possible to create a high, highly optimized algorithm that doesn't render these details when they are not visible. This means a huge performance boost. If we are zoomed out. Two optimizations are introduced so that fiber level self shadows are computed in real time, which would normally be an extremely long process. Note that we are talking millions of fibers here. And three the graphical card in your computer is amazingly effective at computing hundreds of things in parallel. However, its weak point is data transfer, at which it is woefully slow to the point that it is often worth recomputing multiple gigabytes of data right on it just to avoid uploading it to its memory. Again, this algorithm generates the fiber curves directly on the graphical card to minimize such data transfers and hence it maps really effectively to the graphical card. And the result is a remarkable technique that can render a piece of cloth down to the tiniest details with multiple different kinds of yarn models and in real time. What I really like about this piece of work is that this is not a stepping stone. This could be used in many state of the art systems as is right now. The authors also made the cloth models available for easier comparisons in follow up research works. Thanks for watching and for your generous support. See you next time. Ra.