It's surprising that difference is so significant, since I would imagine that the random rays chosen towards the light source might also, by chance, miss the blockage, causing random bright spots in the shadow.
caelinsutch
Have people looked at this problem from a signals perspective - like how quickly the light is changing from "shadow" to "light" and then using that to smooth out sampling?
saif-m17
I thought the point about how, in some cases, it is difficult to be precise in what the light sources are. For instance, in the lecture hall, the projector/ceiling have reflected light, while there is a direct light source that shines light on the projector. I am curious how you might take that into account in sampling and how difficult it is to take into account all the possible light sources.
s3kim2018
Sampling at only the light source area works when you have one light source. I think this will be harder if we have 1. many light sources with different colors or/and 2. One big light source and many small light sources. Since it is a random sample, there could be cases where we only sample from the Red and Green light and not other colors (leading to the wrong texture). Another problem would be a big overarching light dominating all smaller point light sources around the scene.
zeddybot
Question I had about this: In the image on the left, we are uniformly sampling over the solid angles in the hemisphere, but in the right, we are uniformly sampling over the area of the light source. If we truly want these methods to be equal, shouldn't we still sample over the solid angles, but only the ones that correspond to the light source? I think there should be a subtle difference between the two (especially if the light is very wide, with the solid angle approach you would be more likely to sample points near the center of the light source as opposed to the far edges, but with the area approach we are uniformly likely to sample any point on the light, but are now more likely to sample points corresponding to solid angles that are close to the ground, so to speak. Is this actually a relevant different, or does any difference get removed by processing from the algorithm? If it actually does make a difference, does it just cause such small differences in the final result that it doesn't matter?
It's surprising that difference is so significant, since I would imagine that the random rays chosen towards the light source might also, by chance, miss the blockage, causing random bright spots in the shadow.
Have people looked at this problem from a signals perspective - like how quickly the light is changing from "shadow" to "light" and then using that to smooth out sampling?
I thought the point about how, in some cases, it is difficult to be precise in what the light sources are. For instance, in the lecture hall, the projector/ceiling have reflected light, while there is a direct light source that shines light on the projector. I am curious how you might take that into account in sampling and how difficult it is to take into account all the possible light sources.
Sampling at only the light source area works when you have one light source. I think this will be harder if we have 1. many light sources with different colors or/and 2. One big light source and many small light sources. Since it is a random sample, there could be cases where we only sample from the Red and Green light and not other colors (leading to the wrong texture). Another problem would be a big overarching light dominating all smaller point light sources around the scene.
Question I had about this: In the image on the left, we are uniformly sampling over the solid angles in the hemisphere, but in the right, we are uniformly sampling over the area of the light source. If we truly want these methods to be equal, shouldn't we still sample over the solid angles, but only the ones that correspond to the light source? I think there should be a subtle difference between the two (especially if the light is very wide, with the solid angle approach you would be more likely to sample points near the center of the light source as opposed to the far edges, but with the area approach we are uniformly likely to sample any point on the light, but are now more likely to sample points corresponding to solid angles that are close to the ground, so to speak. Is this actually a relevant different, or does any difference get removed by processing from the algorithm? If it actually does make a difference, does it just cause such small differences in the final result that it doesn't matter?