Lecture 13: Global Illumination & Path Tracing (96)
rileylyman184
I think this is a really good example when thinking of Monte Carlo estimation. If you only take one sample, you get a lot of noise, as you can see in the picture. However, you can still see the general shape of the scene through all of that noise. This is because the Monte Carlo estimator calculates the correct value in expectation, so it is expected that even with only one sample lots of these points will contain sort of the right value. For example, notice how towards the bottom of the image, it is all mostly black, with some scattering of very light elements. You can see that in expectation, we get many more shadow pixels than ones which somehow get a huge contribution of light.
I think this is a really good example when thinking of Monte Carlo estimation. If you only take one sample, you get a lot of noise, as you can see in the picture. However, you can still see the general shape of the scene through all of that noise. This is because the Monte Carlo estimator calculates the correct value in expectation, so it is expected that even with only one sample lots of these points will contain sort of the right value. For example, notice how towards the bottom of the image, it is all mostly black, with some scattering of very light elements. You can see that in expectation, we get many more shadow pixels than ones which somehow get a huge contribution of light.