Lecture 13: Global Illumination & Path Tracing (89)
kkoujah
Since path tracing is computationally expensive and requires a significant amount of computing power to produce high-quality images, what are some other solutions to emulate what path tracing can do without the computational power of this technique?
stuurge
Alternatives are discussed in this 2014 paper by Tomas Davidovic:
https://cgg.mff.cuni.cz/~jaroslav/papers/2014-gpult/2014-gpult-paper.pdf
The sections "4.2 Survey of Existing GPU Implementations" and "4.3 Proposed Alternative Implementations" show an overview and alternative algorithms that I think you might be looking for in that question. Path tracing is an inherently huge calculation for any scene, so alternatives explained in the paper try to mitigate rendering issue from their optimizations/approaches.
countermoe
Another alternative to path tracing in relying on information in screen-space, that is, information directly on the screen. Instead of ray-tracing against geometry in the scene, you rely on the depth buffer. This is still an expensive computation, but it's much faster than full path tracing. The downside--because we're relying on information in screen space, as soon as an object is removed from view, there's no way to project that information back. You can easily see this in most modern rasterized games, if you have a large body of water reflecting the environment past it, then occlude the environment. The water will lose the reflection. This technique can also provide an approximation of global illumination as well.
adityasingh7311
I think I'm missing the bigger picture but why do we say we are "partitioning the recursive radiance evaluation"? The direct lighting is not a recursive quantity and I don't think it was being calculated recursively before? So what's different here?
chetan-khanna
Given how computationally expensive path tracing is, I was curious about the feasibility of path tracing in moving scenes (e.g. animation, or even video games in real-time). I found some information online about it (e.g., see this link http://gamma.cs.unc.edu/BVH/). I was curious, however, as to whether caching was a possible option here: could we cache certain parts of a scene (which are relatively static, and perhaps aren't receiving much light from moving parts) so that they don't have to be re-computed repeatedly?
pcg108
@adityasingh7311 I think that term is referring to splitting the radiance evaluation within each recursive call. So you don't recurse on the direct lighting, and you do recurse on the indirect lighting
Since path tracing is computationally expensive and requires a significant amount of computing power to produce high-quality images, what are some other solutions to emulate what path tracing can do without the computational power of this technique?
Alternatives are discussed in this 2014 paper by Tomas Davidovic: https://cgg.mff.cuni.cz/~jaroslav/papers/2014-gpult/2014-gpult-paper.pdf
The sections "4.2 Survey of Existing GPU Implementations" and "4.3 Proposed Alternative Implementations" show an overview and alternative algorithms that I think you might be looking for in that question. Path tracing is an inherently huge calculation for any scene, so alternatives explained in the paper try to mitigate rendering issue from their optimizations/approaches.
Another alternative to path tracing in relying on information in screen-space, that is, information directly on the screen. Instead of ray-tracing against geometry in the scene, you rely on the depth buffer. This is still an expensive computation, but it's much faster than full path tracing. The downside--because we're relying on information in screen space, as soon as an object is removed from view, there's no way to project that information back. You can easily see this in most modern rasterized games, if you have a large body of water reflecting the environment past it, then occlude the environment. The water will lose the reflection. This technique can also provide an approximation of global illumination as well.
I think I'm missing the bigger picture but why do we say we are "partitioning the recursive radiance evaluation"? The direct lighting is not a recursive quantity and I don't think it was being calculated recursively before? So what's different here?
Given how computationally expensive path tracing is, I was curious about the feasibility of path tracing in moving scenes (e.g. animation, or even video games in real-time). I found some information online about it (e.g., see this link http://gamma.cs.unc.edu/BVH/). I was curious, however, as to whether caching was a possible option here: could we cache certain parts of a scene (which are relatively static, and perhaps aren't receiving much light from moving parts) so that they don't have to be re-computed repeatedly?
@adityasingh7311 I think that term is referring to splitting the radiance evaluation within each recursive call. So you don't recurse on the direct lighting, and you do recurse on the indirect lighting