Continuing from last post
, DDGI requires a high amount of memory to store its depth maps. I present a compression scheme that compresses the data to 5~10% of its original size. Note this is completely orthogonal to BC texture compression or any bit-saving tricks, so this compression ratio can be improved further with those existing techniques.
For DDGI to work effectively, we need a 16x16 depth map attached to each probe, with each texel storing both depth and depth^2. If we use R16G16 to store each texel, then that would give us 1KB per probe, not counting ...
Read More →