My educational guess would be supersampling and averaging down triangles one by one would be much more memory efficient than projecting all triangles onto a high-res gird first. The reason being in the former case computer could release memory for high-res grid after each triangle is processed while the latter case computer needs to hold memory for all high-res grids until all triangles are processed.
dstnluong
Why is it called "average down" rather than just "average"?
JefferyYC
After taking the average the number of samples in a pixel box decreases from 4 to 1.
RK900
Performing this method one-by-one for each triangle seems much more memory-friendly, since you can just free up space after supersampling+averaging down each triangle.
However, how would this hold up if you have adjacent triangles? If 2+ triangles share the same border, would it make more sense to render them both at the same time?
davidmwei
I had a similar question to @RK900: It seems simple to sample when you have only one triangle, or even multiple triangles which are spaced apart. How do you sample when you have many triangles sharing the same border? Do you average down and then take the average value of the border (wouldn't this lead to a problem to very inaccurate borders?)? Or do you have to "know" to strictly supersample at certain locations.
xgao0613
I was wondering if there is anything special we need to do when super-sampling multiple triangles overlapping on top of each other? In that case do we super-sample each triangle or each sub-region?
My educational guess would be supersampling and averaging down triangles one by one would be much more memory efficient than projecting all triangles onto a high-res gird first. The reason being in the former case computer could release memory for high-res grid after each triangle is processed while the latter case computer needs to hold memory for all high-res grids until all triangles are processed.
Why is it called "average down" rather than just "average"?
After taking the average the number of samples in a pixel box decreases from 4 to 1.
Performing this method one-by-one for each triangle seems much more memory-friendly, since you can just free up space after supersampling+averaging down each triangle.
However, how would this hold up if you have adjacent triangles? If 2+ triangles share the same border, would it make more sense to render them both at the same time?
I had a similar question to @RK900: It seems simple to sample when you have only one triangle, or even multiple triangles which are spaced apart. How do you sample when you have many triangles sharing the same border? Do you average down and then take the average value of the border (wouldn't this lead to a problem to very inaccurate borders?)? Or do you have to "know" to strictly supersample at certain locations.
I was wondering if there is anything special we need to do when super-sampling multiple triangles overlapping on top of each other? In that case do we super-sample each triangle or each sub-region?