[MEncoder-users] Fwd: Questions on Mencoder performace

[Phil Ehrens]

Jeff Clagg wrote:
> On Wed, Jun 07, 2006 at 08:40:43AM -0700, Phil Ehrens wrote:
> > Strange. I am getting no blockiness at all. I am using a target
> > bitrate of 2400 kbps and seeing actual average bitrates as low
> > as 800 kbps for very noisy material. For the very clean material
> > with high action the average bitrate is 2400 kbps, and there is
> > no blocking at all. Again, this is for 2-pass encodes. For single
> > pass the image degrades rapidly above nr=200.
> > 
> > I wonder why we are seeing such different behaviour? Maybe nr
> > interacts badly with some other options or filters? My tests
> > are all dvd -> lavc mpeg4 with no resizing, just cropping.
> 
> Quick uneducated guess: nr has way less practical effect at higher
> quantizers, since higher frequency coefficients were less likely to be
> coded anyway (and that is where the noise is).

This makes sense if nr is, as I suspect, a filter and decimate
operation. I have a very poor conceptual grasp of what a higher
quantizer is, however. Commercial dvd's seem to have a large
amount of added stochastic noise which is not normally encoded,
and using nr=1000 seems to have the effect of making the stochastic
noise rejection a little bit more aggressive, but not to the point
of producing noticeable artifacts when this high frequency noise
component exists. If one operates on source material without added
stochastic noise, then I suppose nr will merrily go and attack the
signal, producing the blocking that other people are observing.