[]
if you visualise the cascade frustums, they all are nested inside eachother and just move depending on the viewing direction - that’s pretty bad yo.
[/]
[]
regardless of the viewing angle the cascades never leave eachother’s bounds, the largest bound contains all of the smaller ones completely
[/]
That’s not the case, larger cascades are not placed to contain smaller cascades. However, with a large camera FOV which is typical, the texture mapped to the larger cascade can end up encompassing the smaller cascade. Object culling on the CPU is still using the non-overlapping bounds of the cascade. Objects only have to be drawn into multiple cascades if they are actually affected by multiple cascade depth ranges. Have a look at GetShadowSplitBounds().
The image you posted looks like clipmaps which aren’t view direction dependent at all, unlike cascades, so it’s completely unrelated.