[x264-devel] Re: patch: switch to select ME method & range

[Tuukka Toivonen]

On Thu, 17 Mar 2005, Måns Rullgård wrote:

>> encoding quality reducing bitrate at constant quality in average 4%
>> [1], at low bitrates, making x264 pretty much on par with JM. It's
>> just a bit slow.
> Slower than JM?  That's, well... slow.

Not _that_ slow [1] :)
Let's see: ESA is "only" 3-4 times slower than hexagonal search.

> IMHO, QP 36 is barely usable, except, possibly, if you need to
> transfer video the size of a postage stamp from some cheap web-cam
> over a 14.4 kbps modem link, and, for some inexplicable reason, are
> satisfied with that.

Yeah, and I suppose you imply that if one is satisfied with that quality he 
doesn't need that extra quality from ESA.

Well, generally ESA helps also at higher quality, just not so much:
http://www.ee.oulu.fi/%7Etuukkat/tmp/x264-r176-me.patch-cifresults.eps
http://www.ee.oulu.fi/%7Etuukkat/tmp/x264-r176-me.patch-cifresults-stefan.eps
The first is average over all 6 sequences.

[1] Actually the first try was to loop over whole frame when searching a 
match for a block but that was too slow even for me ;) So I had to put some
limits there... or maybe I should just have demanded assembly programmers 
write a faster SAD routine?-)