Code: Select all
Wrong threadrandom hemisphere (tlrcam related
kram:
collumbia, cornell, mit, caltech, nyu, brandise, carnage mellon, georgia tech, bu, rit, gorge washington, university of washington, stonybrook....
CTZn:
my engine is now unbiased and progressive... rays are computed using the surface normal, but they still need to be in random directions. the diffuse shader throws a random ray on the hemisphere in the direction of normal.
collumbia, cornell, mit, caltech, nyu, brandise, carnage mellon, georgia tech, bu, rit, gorge washington, university of washington, stonybrook....
CTZn:
my engine is now unbiased and progressive... rays are computed using the surface normal, but they still need to be in random directions. the diffuse shader throws a random ray on the hemisphere in the direction of normal.
a shiny monkey is a happy monkey
Another question, to see if I can get it right:
You mean the normal of surfaces determines the hemisphere position and orientation, while rays are shot randomly from the point on surface through the hemisphere ?
Well, makes sense, I saw that somewhere already... it's a classical algorithm isn't it (I mean old school) ?
You mean the normal of surfaces determines the hemisphere position and orientation, while rays are shot randomly from the point on surface through the hemisphere ?
Well, makes sense, I saw that somewhere already... it's a classical algorithm isn't it (I mean old school) ?
obsolete asset
yah- you got it right.
i already realized what the issue was with my old algorithm. the get random points on a disk was weighted towards the center of the disk. it worked by getting a random ray on a unit circle around a vector and scaling it to a random size. think about a point on an infinite plane. now in your mind try to fill up the plane completely with lines passing thorough the plane. its mentally impossible because the solution is an infinite number. at that infinite number the concentration of the distribution lines grows larger closer to the point of origin. the same goes for filling up a unit sized circle with lines passing through the center of the circle. with a fixed number of lines, if a same sized sample is taken at any point in the circle containing the lines, there will be more lines in that sample when taken near the center then there will be near the edge. because of this, when a random points are taken on these lines, there will be a higher chance that the random point is nearer the center because there are more lines near the center. at least, thats what i think would be wrong, it could also be that the random selection was cylindrical forced into a hemisphere. probably the latter because the former still confuses me a bit.
i already realized what the issue was with my old algorithm. the get random points on a disk was weighted towards the center of the disk. it worked by getting a random ray on a unit circle around a vector and scaling it to a random size. think about a point on an infinite plane. now in your mind try to fill up the plane completely with lines passing thorough the plane. its mentally impossible because the solution is an infinite number. at that infinite number the concentration of the distribution lines grows larger closer to the point of origin. the same goes for filling up a unit sized circle with lines passing through the center of the circle. with a fixed number of lines, if a same sized sample is taken at any point in the circle containing the lines, there will be more lines in that sample when taken near the center then there will be near the edge. because of this, when a random points are taken on these lines, there will be a higher chance that the random point is nearer the center because there are more lines near the center. at least, thats what i think would be wrong, it could also be that the random selection was cylindrical forced into a hemisphere. probably the latter because the former still confuses me a bit.
a shiny monkey is a happy monkey
Who is online
Users browsing this forum: No registered users and 61 guests