Ancestors

Toot

Written by cnlohr on 2024-12-26 at 09:53

About 6 months ago, Lyuma2D told me about Arithmetic coding, saying it was "Better than Huffman" which I laughed at, because, Huffman is provably optimal (but only for symbols). Boy was I wrong. I extracted VPX (VP8/9) coding into a single-file library https://github.com/cnlohr/vpxcoding

=> More informations about this toot | More toots from cnlohr@chaos.social

Descendants

Written by cnlohr on 2024-12-26 at 11:10

This library is still very work-in-progress, but it's what I'm using in another project, and indeed it is able to eek out between 1-5% improved overall compression compared to huffman coding, in almost all the cases I've been able to find. More importantly in many cases sometimes it's easier to use!

=> More informations about this toot | More toots from cnlohr@chaos.social

Written by Giles Goat on 2024-12-26 at 13:01

@cnlohr I remember YEARS ago reading something somewhere about it and "the reason it wasn't much used" ( so the article at that time said ) was some patent issue that was preventing others to use it ?

=> More informations about this toot | More toots from gilesgoat@toot.wales

Written by 🇺🇦 haxadecimal on 2024-12-26 at 13:15

@gilesgoat @cnlohr Yes, there were patents on arithmetic coding. AFAICT, they are now expired, but don't take my word for it.

=> More informations about this toot | More toots from brouhaha@mastodon.social

Written by 🇺🇦 haxadecimal on 2024-12-26 at 13:19

@gilesgoat @cnlohr

In practice, most newer compression standards (e.g. Zstd), instead of using arithmetic coding, instead use asymmetric numeral systems. ANS has essentially the same compression ratio as AC, but at lower computational cost.

=> More informations about this toot | More toots from brouhaha@mastodon.social