How would a CS degree help with any of that? I'm someone with experience in weird uber-specific, low-level areas of computer graphics/imaging with no CS degree, and none of the CS graduates I know have been taught those things at university.

Some PhDs and math students yes (at least, with regards to the algorithmic portions), but otherwise you have to learn the low-level esoteric stuff on your own, or be lucky enough to have a specialized optional course at your university.

And the cutting-edge will rarely be available to undergraduates anyway. All my CS graduate colleagues learned about codecs was how to compute a DCT by hand - they certainly didn't learn the modern 4x4 and 8x8 spatial block transform approximations as used in h264, or any motion vector calculation algorithms, or the x86_64 SIMD intrinsics, or any of the modern entropy coding methods.