This is another one of the Global Illumination Test Scenes -FROM THE YEAR 2000-presented here.
Call this one 5 out of 6.
Here’s a 1 sample per pixel render of this scene for discussion. There’s a big light source above the image shining down on the big grey square which casts a shadow at the bottom of the image.
Below the big grey square, there’s another, smaller light source, and another, smaller grey square.
1 sample per pixel, taking a look at the problem.
That sample image shows us the frame we’ll be rendering, but to see the larger light source, we have to back waaay up.
The wide wide wide view of the problem.
That larger light source might not look big from here, but it’s 100 times bigger than the other one. They called this scene huge.obj.
If you’re using importance sampling based on size alone, the huge light source can trip you up, because you might end up sampling it way more than the small light source because of its size, even though the smaller light source is actually way more important in our image.
For example, in the large shadow area, the smaller light source is really the only important light. So it’s a shame if 100/101 samples get spent on a big light source where it’s shadowed and contributes nothing.
Photon mapping is a neat way to handle cases like this, but I’m not there yet.
Thankfully, since the huge light source is so far away, my importance sampling already balances out this scenario. It also doesn’t hurt that I rendered this scene with 100 samples per pixel, and 16 light samples, so the result is nice and smooth.
100 samples per pixel, no indirect.