Log in

View Full Version : Alt.CC options are explained here.


OUTPinged_
7th June 2002, 18:42
Here it comes, simple stuff followed by more complex. User-friendly explanations are in italic.

Automatic minimum relative quality - a simple coefficient for calculating min_rel_qual variable. It automatically looks at the first to second pass ratio (GKnot users are familiar with it) and multiplies it by Strength percentage. The result is capped between 20 and 100.

minimum_relative_quality=100-(100-encode_quality)*strength<--like that

Wicked. All numbers in italic section are to be changed.

ie 1/2pass=66, strength=50, mrq will be equal to 100-(100-66)*0.50=83.


Very newbie-friendly option. The working range is 50-100, with 90-100 for 2cd rips (66%+ 1/2pass ratio) and 50-65 for 1cdrips of long movies (35-50% 1/2pass)



Bonus bias - determines how much bit (percentage) will be distributed "proportionally" and by "bias". 100 equals pure bias distro and 0 equals to pure proportional distro of bonuses. 50 gives equal amount of bits to both.

"Proportional" gives more constant quality throughout the movie while "bias" will give more bits to smaller (lo-motion) frames.


Automatic bonus bias - sets "bonus bias" equal to min_rel_qual variable value.

That one is nice. Gives pretty balanced results and one option less to get headache with. Will favor lo-motion scenes for high quality encodes and give more steady quality for 1CD rips

Minimum relative quality - a coefficient (it's a percentage, ie 33mrq is 0.33 coeff.) by which alt.cc algo multiplies *all* frame sizes *except* frames that are so small that dont fit into "low distance" range.

Ok, this one is pretty important. If you are setting it manually, you need to have some understanding about quantizers and know how high quant. values do you want to reach on high-motion scenes. To make things more simple: if 1/2pass ratio is 66%(av. quant.=3) and mrq=50, your maximum quantizer will not rise higher than quant6, even on _very_ intense scene. It is a bad idea to get out of 40-66 range of this variable, and 50 seems to me like a value close to ideal. This setting is the most responsible if your movie will look like "MAED iN FL4SK!!11(c)" or not.

High/low distance - defines a part of framesize range, which will be multiplied by "aggression" coefficient. Ranges begin at average_framesize (0), "low distance" ends at 100 (with framesize=0), "high distance" is uncapped.


It is surprisingly hard to make an encode look terrible with these, and tweaking them wont help at all if you have set MRQ wrong.
If you did not, then setting low dist. to 25-50 and high dist. to 100 with high agg. and 200 with medium and low agg. will typically yeld pretty decent results. For more "old curve compression" feel, set low. dist. to 100 or higher(codec will set it back to 100 in that case)


Aggression - a coeficient by which all frames with framesizes that are within "distances" multiplied. "Medium" is a constant multiplier and "high" and "low" are sine/cosine multipliers.

This is mostly a matter of personal preference until you are _really_ knowlegable. Begin with "medium" and when you will be getting stable results(means other settings are ok), try to change it to high/low. That change will affect quite much frames, with hih agg making "medium intensity" scenes look a tad worse(with increase of quality in hi-mo scenes) and low agg. will do the opposite.


And a hint: dont touch those "Max overflow improvement/degradation" options unless you know what they will do. They are not "max bitrate" related (are they?).



PS.I will answer questions here and will be happy if Foxer will point out if i was wrong at the descrptions somewhere.

[edit] typos atm

-------MooPolice Approved \o/

Foxer
7th June 2002, 20:58
Pretty decent explanation though a few things are explained as the opposite of what they are.

"ie 1/2pass=66, strength=50, mrq will be equal to 66*0.50=33"
would be
ie 1/2pass=66, strength=50, mrq will be equal to 100-(100-66)*0.50=83

"Very newbie-friendly option. If you are not capping quantizers, set it to at least 50% (25-30% are a no-no). for nice 2cd rips values can be in 75-90 range."

Are you talking about mrq or automatic mrq here because 25-30% strength with automatic mrq would have less effect on the quality variation than 75-90%

The codec doesn't just set the low distance back to 100% if you specify over 100% but rather modifies the internal variables so that 100% to mrq is still attainable using the lower shape defined by the low distance & specified aggression.

It appears you like to be a tad extreme with high aggression distances heh

High aggression will improve/degrade low/high bitrate scenes the fastest and low aggression will improve/degrade low/high bitrate scenes the slowest.

Max overflow improvement/degradation aren't max bitrate related, no.

OUTPinged_
7th June 2002, 22:04
1/2pass=66, strength=50, mrq will be equal to 100-(100-66)*0.50=83

My mistake. Lost one "(" while browsing source.




Editing....

unplugged
7th June 2002, 22:57
Originally posted by Foxer
Are you talking about mrq or automatic mrq here because 25-30% strength with automatic mrq would have less effect on the quality variation than 75-90%

So, VBR purist will choose 100% as Strength for MRQ?
I think yes, this seems pretty clear.

Originally posted by Foxer
The codec doesn't just set the low distance back to 100% if you specify over 100% but rather modifies the internal variables so that 100% to mrq is still attainable using the lower shape defined by the low distance & specified aggression.

Excuse me, Foxer, could you explain a bit more widely this interesting argumentation?
Would be greatly useful, not only for me (hope)
Thanks :cool:

Foxer
7th June 2002, 23:16
@unplugged

A VBR purist would set automatic mrq's strength to zero or mrq to 100%

Since a negative threshold (low distance of greater than 100%) is unreachable, if the low distance is larger than 100%, it cuts it off at that point and uses only that part of the lower curve. This is shown in the graph I scratched up.

unplugged
8th June 2002, 00:35
Does Setting MRQ (not MRQ strength) below 100% starts to decenter (down) the average frame size?

Sorry :o

OUTPinged_
8th June 2002, 11:49
No, it does the opposite.

The higher is mrq, the less curve compression will be applied.


Basically there are 3 frame ranges that count:

1. Very small frames that fall under low dist (there are none when low dist>99) - are set to average quantizer and then grow only by bonus distroing. Usually there are frames that contain stills or very dark/simple lo-mo scenes.

2. Framesizes that fall into high/low distance. Usually more than 80% of all frames fall into that category. Are affected by aggression, mrq and bonuses. Most of the "lo-motion" scenes are there, and quite a bit of "hi-mo", too.

3. Frames that are bigger than high distance. Are affected by mrq and bonuses. Here we have intense "hi-mo" scenes, high-detail pannings and unusually large frames(ie artificial noise). Basically if your hi-motion scenes look like crap, mrq is set too low.

Tbh, quantizer distribution is not that big deal, and 25low/200high with low agg. and 50(2cdrips)-70(1cdrips)mrq should produce pretty good results no matter what the encode is.

unplugged
11th June 2002, 19:01
MRQ is a coefficient, but if it acts as curve compression factor then it has the average-frame-size as center bias. I'll explain better my thought

for example if it is 80%:
frames above average frame size get cut by 20% of the upper distance from average frame size
frames below average frame size get gain by 20% of the lower distance from average frame size

So if it works this way, why has it so strange name? :) (Minimum Relative Quality) To what Minimum?

again,
?


Feel free to not re-post, or post so shortly to not influence your Minimum Relative Time... :) thanx

OUTPinged_
13th June 2002, 08:53
MRQ defines the maximum amount of bits that will be removed by the algo.

In your case even the most largest frames will be scaled to av. quantizer and then have 20% of bitrate shaved off.


Minimum relative quality - "minimum frame quality relative to average frame quality" or something like that :-)