[REQ] Go with the Winners Strategy in Path Tracing

[Zom-B]

Abstract:

This paper proposes a new random walk strategy that minimizes the variance of the estimate using statistical estimations of local and global features of the scene. Based on the local and global properties, the algorithm decides at each point whether a Russian-roulette like random termination is worth performing, or on the contrary, we should split the path into several child paths. In this sense the algorithm is similar to the go-with-the-winners strategy invented in general Monte Carlo context. However, instead of establishing thresholds to make decisions, we compute the number of child paths on a continuous level and show that Russian roulette can be interpreted as a kind of splitting using fractional number of children. The new method is built into a path tracing algorithm, and a minimum cost heuristic is proposed for choosing the number of reflected rays. Comparing it with the classical path tracing approach we concluded that the new method reduced the variance significantly.

http://www.iit.bme.hu/~szirmay/gowin2.pdf

[Deus]

Girls always go with the losers. Ill stick to that.

(Interesting paper though )

In the end... brute force and more cores always wins

[Robotbeat]

I don't think brute force always wins. In fact, algorithmic efficiency for many things has actually increased FASTER than Moore's Law. That means that for many classes of algorithms, if you tried today's state-of-the-art algorithm on hardware thirty years old, it would actually beat out the thirty-year-old algorithm running on modern machines!

That said, I think that most programmers outside of math and science have gotten more lazy, so overall programming efficiency may well have gone down in non-math and non-science fields.

[Deus]

Im a professional software developer. I was trying to be funny, hence the smiley which I have been told by moderators that I should add to every sarcastic comment.

To be serious the flip side of the coin. Most really nice algorithms are not feasible without more hardware. Raytracing was pretty unreasonable in terms of speed and quality before FPUs, and also to be able to develop, design and test algorithms the development cycles needs to be short. Back in the days when I wrote my first raytracer a row of pixels easily took minutes on moderate complexity scenes with trivial lighting.

The prettiest raytracing algorithms is yet to come.