Yeah - I thought I was doing fairly well back in the 70s when I wrote a (lossless) half-tone compression routine that got ~50% compression when there was only enough core [and some of the machines it ran on uses actual core] to hold one scan line. This stuff is amazing.