I wonder if there is a way to dynamically scale grid resolution over a scene so that less computation needs to be done in areas where there are less objects. It seems that the C value in 3D should also be very dependent on the complexity of shapes being rendered as well.
arjunsrinivasan1997
Is this the only heuristic that is used when trying to calculate the grid resolution, or are the other potential heuristics that have niche applications?
fywu85
The number 27 seems to be a little arbitrary to me. Is it the number of elements inside a 3 by 3 by 3 cube? If so, why? In case of 2D, is C equal to 9?
gprechter
I agree that the value of C as 27 is a little arbitrary. One thing it does remind me of is the dimensions of a convolutional kernel, used in image processing, which is often a 3 x 3 grid in a 2D image. The best explanation I've found for that number is that it can easily be centered around a single pixel. Perhaps it's best to use a 3 x 3 x 3 cube, for each object so that the grid can easily be centered around an arbitrary point? I too would like to know how that number was achieved, is it just a magic number that has yielded the best results in practice?
I wonder if there is a way to dynamically scale grid resolution over a scene so that less computation needs to be done in areas where there are less objects. It seems that the C value in 3D should also be very dependent on the complexity of shapes being rendered as well.
Is this the only heuristic that is used when trying to calculate the grid resolution, or are the other potential heuristics that have niche applications?
The number 27 seems to be a little arbitrary to me. Is it the number of elements inside a 3 by 3 by 3 cube? If so, why? In case of 2D, is C equal to 9?
I agree that the value of C as 27 is a little arbitrary. One thing it does remind me of is the dimensions of a convolutional kernel, used in image processing, which is often a 3 x 3 grid in a 2D image. The best explanation I've found for that number is that it can easily be centered around a single pixel. Perhaps it's best to use a 3 x 3 x 3 cube, for each object so that the grid can easily be centered around an arbitrary point? I too would like to know how that number was achieved, is it just a magic number that has yielded the best results in practice?
https://www.cs.utah.edu/~aek/research/grid.pdf