Introduction to Computer Graphics (Lecture 16): Global illumination; irradiance/photon maps

6.837: Introduction to Computer Graphics
Autumn 2020
Many slides courtesy past instructors of 6.837, notably Fredo Durand and Barbara Cutler.

[@sungjuyea4627]

Very Informative, and way clearer in explaining this huge and difficult subject than any other materials that I have met before. Thank you so much!

[@Jonasz314]

Solid video with clear explanation of advanced concepts. One thing that should be mentioned for all these advanced techniques (MC pathtracing, Photon Mapping, Irradiance Caching...): they converge VERY, VERY SLOWLY. I think you already show this in a couple of slides, but you could probably harp on a big more on the massive cost of trying to compute global illumination. With pure, basic raytracing you will get a crisp image with incorrect data but ~no noise, an image will render in seconds assuming you have decent traversal routines for complex objects (e.g hierarchical regular grid for meshes). The minute you try to slap global illumination on your solution, prepare for a massive change. At first the images you get will look downright ugly and totally noisy. You need to collect a TON of sample per pixel to get something relatively smooth. Computation time jumps from seconds to minutes or hours per frame for a Cornell box, so anything more complex than this will be even worse.

[@germolinal]

Radiance is not only "still there" but it is widely used for performing Daylight analysis in Buildings.

[@32zim32]

Amazing. Thanks

[@kendiato8714]

thank you very much for the video!

[@d7ffab979]

Why don't they used compressed sensing to only sample a fraction of the points and reconstruct the whole signal from it.

[@ringo8530]

11:24 35:41 55:02

[@d7ffab979]

Why don't you construct the solution as a fixpoint in a banachspace?

[@pifre3051]

I haven't even learned caculus.

[@iloveyourwcw9265]

Take baby steps bruh