[MPlayer-users] Suggested Low Bitrate Encoding Options

[Corey Hickey]

RC wrote:
> On Sun, 26 Aug 2007 12:15:40 -0700
> Corey Hickey <bugfood-ml at fatooh.org> wrote:
> 
>> Try vrc_eq="(tex+10^8*mcVar)^0.6". That's my current favorite, and, 
>> though it isn't particularly tuned, it works well enough for me. 
>> High-motion scenes should look much better and low-motion scenes
>> should  look very slightly worse (that's the tradeoff).
> 
> With my test video, high motion scenes don't look any better at all, and
> low motion scenes look much, much worse.

That surprises me. I'm curious to try encoding your test clip. Can you 
post it somewhere, or make a representative sample that is large enough?

This might motivate me to dig deeper into the ratecontrol 
experimentation I started. I don't have any time in the short term, but 
I should be able to in a few weeks.

> Same with/without vqcomp=0.7.

Right. vqcomp is referred to in the ratecontrol equation as "qComp", 
such as in the default, "tex^qComp". If you specify an equation without 
qComp, the vqcomp parameter won't affect anything.

> Perhaps it just doesn't mix well with my other options (*_mask)?

I don't know. I wouldn't expect it to conflict, but I don't use any of 
the "psychovisual" options, so I suppose there is a possibility.

> For your suggestions from the other thread:
> 
> *cmp=2 makes my clip _far_ more blocky in several places (where-ever it
> was slightly blocky without *cmp=2 to begin with), it makes the rest
> just slightly more blocky than normal.  It certainly doesn't improve
> quality, anywhere.
> 
> qns=1 looks slightly worse/blockier throughout. 
> qns=2 has some odd effects, discoloring the occasional macroblock, and
> slightly changing some artifacts (to different artifacts).  In any case
> it really doesn't improve anything.

Again, I'm curious to try your test clip, especially with the other 
parameters you're using.

> I've tried vb_strategy=1 with the full range of values. b_sensitivity=5
> seems to be the best of them, but it's still doesn't do nearly as well
> as vb_strategy=2.  I guess the extra encoding time is worthwhile.
> 
> sc_factor=6 has nominal effect.  Perhaps just barely lower quality, if
> anything.

Er... I think it's your imagination, unless your mencoder is rather old. 
sc_factor doesn't do anything as of ffmpeg r6876. I have been remiss in 
keeping up with such things lately.

http://svn.mplayerhq.hu/ffmpeg?view=rev&revision=6876
http://svn.mplayerhq.hu/ffmpeg/trunk/libavcodec/motion_est.c?r1=6829&r2=6876&pathrev=6876

Don't worry, the same thing has happened to me before.

-Corey