So, unlike those breakthrough codecs everybody talks about (I mean RMHD and ORBX.js), V-Nova Perseus was delivered (but what do you expect from a codec announced on the first of April?) and is available in some Android app. So I’ve looked at it.
The implementation seems bafflingly simple: there’s a base layer, it gets upscaled 2x and an enhancement is applied to the upscaled image. And those enhancements are essentially quantised differences after 2×2 Haar transform plus runs, all coded with context-dependent Huffman codes. If that reminds you of RealVideoâ€”don’t worry, they code those codebook descriptions too so it’s different.
I don’t know if it really works as good as
promisedmarketed but it’s an interesting approach and it introduces some variety in the world of codecs that look alikeâ€”mostly because they all use the same principles as the standard video codec with some small enhancements or building blocks replaced with functional analogues; yes, I completely forgot about Daala, please remind me about it when they settle with final designâ€”it might be the codec of choice for GNU HURD NG by then too.