[MEncoder-users] Tuning hqdn3d

[vmrsss]

On 13 May 2009, at 09:16, RC wrote:
> The only difference being that CRF will only consider it's
> best-guess for quality, while 2-pass mode will chose the best possible
> quality for a given bitrate distributed over the length of the file,
> within specified min/max instantaneous bitrates.

yet, this is non-sense: best-guess for quality what if you're not  
targeting any specific bitrate but just specifying directly the  
quality frame by frame. There is no guessing involved (the same  
equations are in action in the two cases, in one you must derive a  
rate factor from bitrate and the rest, in the other you're given the  
rate factor). Possibly you are trying to say that it is guesswork for  
an encoder to establish a reliable correspondence between a required  
visual quality and a CRF value?; that's very true, but then the same  
is true for picking a target bitrate, isn't it? (admittedly I am  
ignoring vbv and max bitrates)

> But more importantly, and less controvercially, some options just
> require look-ahead, which x264 in single-pass/CRF mode simply can't  
> do.

That's an interesting point. I remember a discussion on doom9 that in  
fact only a very short (a few frames) lookahead really affects the  
encoding, but that will surely depend on what you're trying to  
optimise for. It should be possible to come up with meaningful  
(essentially) global requirements, such as the one in the previous  
post about the TeX line splitting algorithm applied to optimise scene  
changes.