1-pass QB or Multipass for best quality

[mxclouti]

I want the best quality possible with DIVX 5.0.3. The size of the AVI does not matter to me.

Will I get it with

1. 1-pass quality-based, quantizer = 2, no qpel, no GMC, no b-frames

or

2. Multipass (2 passes), encoding bitrate = 16000 kbps, no qpel, no GMC, no-frames.

-----------

Also, I encoded a movie with 5.0.2 this time, 2-pass with an encoding bitrate of 5000. I then encoded it with 8000 and 10000. Every time I get about the same size AVI file. Does the mean that at 5000 (maybe less) I achived the maximum quality possible?

[BoNz1]

1. Will always give better quality. Use it if you don't care about filesize. It will waste bits, but that doesn't matter since you don't care about filesize.

To answer your second question. It is simple, you saturated the codec. It doesn't need that high bitrate if it did use it what you told it to, it would be a waste. :)

[N_F]

Option 1 is your best choice.

Originally posted by mxclouti
Also, I encoded a movie with 5.0.2 this time, 2-pass with an encoding bitrate of 5000. I then encoded it with 8000 and 10000. Every time I get about the same size AVI file. Does the mean that at 5000 (maybe less) I achived the maximum quality possible?

It's not quite as easy as that, but basically, yeah.

The number is usually much lower than 5000. For movies that compress really well it may be 900-1000. "Normal" movies 1700-2000. Extrem movies 3000-4000.

But this also depends a lot on what resolution you're encoding at and what filters you're using.

[N_F]

Originally posted by BoNz1
To answer your second question. It is simple, you saturated the codec. It doesn't need that high bitrate if it did use it what you told it to, it would be a waste. :)

Strangely DivX 5.03 doesn't seem to behave this way. On some tests I ran some days ago I managed to get a considerably larger filesize with very high bitrates than with quant 2. :confused:

I haven't looked more into it since then, perhaps I should...

[jonny]

Strangely DivX 5.03 doesn't seem to behave this way. On some tests I ran some days ago I managed to get a considerably larger filesize with very high bitrates than with quant 2.

One of my friends told me that the 5.0.3 uses quant=1 sometimes ... time to test :)?

I've used 1 clip of 1000 frames.
(same parameters for all the encodes, only the mode change)

1 - original 2-pass, br=16000: file size = 21,413,888 bytes
2 - multi-pass (2 pass), br=16000: file size = 24,752,128 bytes
3 - quality based, quant=2: file size = 9,320,448 bytes

Now the interesting part: ffdshow show almost all the quant=1 in the case 1 and 2!

(image attached)

[BoNz1]

Very interesting. I knew that divx5.03 can use quant 1, but I thought that you had to tell it that it could use it, ie from the CLI when you specify the max and min quants. Since, codec always seems to erase whatever I put in there I never bothered to find out whether it was actually using it or not since I thought the max and min quants where the same defaults from 5.02 2 and 12. Thanks for the info, jonny!

[jonny]

The bad thing is that there is no way to control this... divx.com have "forgot" to put RC parameters controls.
One thing that i can't really understand, at this point, it's why the minimum quant isn't =1 in quality based mode.
:angry::confused::angry:

[jonny]

Some more tests:

Original 1-pass, br=16000: file size = 9,287,680 bytes
1-pass, br=16000: file size = 23,019,520 bytes

Seems that quant=1 is not used in "Original 1-pass".

Anyway the strangest case is "Original 2-pass"...
In this case i've obtained a little amount of quants=2

Probably there are some quants=2 in 1-pass too, because the file size is a bit little, compared to multi-pass.

[DJ Alik]

what about writing -b1q 1 in the CLI . wouldn't that work? i haven't actually tried that out but i will tonight

[BoNz1]

Yes, you can specify the max min quant as well as the rc averaging period from the CLI as I said before. However, like I said everytime I try to put anything in there it erases it. This is a known bug with divx5.03.
EDIT: grammatical errors ;)

[jonny]

-b1q 1 is keeped in the interface, i haven't made an encoding test but i think should work.

[DJ Alik]

it worked. i tried it out last night. i am gonna see how the quality compares to quant2

[mxclouti]

Have a look at the registry: HKEY_CURRENT_USER\software\divxnetworks\divx4windows

With 5.0.2 will see the following keys:

RC Period = 2000
RC Reaction Period = 10
RC Reaction Ratio = 20
Max Quantizer = 12
Min Quantizer = 2

You won't see them in 5.0.3.

If you change the Quantizer in 5.0.3, the key "VBR Quality v5" is updated.

So, there must be a big different between 1-pass QB 5.0.3 vs 1-pass QB 5.0.2?

See my post " 1-pass QB 5.0.3 vs 1-pass QB 5.0.2 .. strange, some answers are needed"

[jonny]

RC Period = 2000
RC Reaction Period = 10
RC Reaction Ratio = 20
Max Quantizer = 12
Min Quantizer = 2

afaik those option are not used in 1pass qb, why do you care about this?

If you change the Quantizer in 5.0.3, the key "VBR Quality v5" is updated.

DivX 5.0.2 should have a similar key for the quality based quant, this value is not related to the RC parameters.

[mxclouti]

In DIVX 5.0.2, for any Bitrate mode (Bitrate tab), all Data Rate Control (RC) parameters (in the Advanced Parameters tab) are accessible. So they must be used in all modes?

Why in 5.0.3 we don't have access to the Data Rate Control (RC) parameters? They are not present in the registry too.

[jonny]

Do you think changing the minquant or maxquant on a quality based encode (quality based encode = fixed quant) have some effect?

We all know that we can't set RC parameters in 5.0.3 ... then?

I don't get the point of your discussion.

If you want to say that the RC parameters make big differences in quality based mode, simply do your tests with 5.0.2 changing those values.


In this thread: http://forum.doom9.org/showthread.php?s=&threadid=46265
gldblade asked you some clarifications ...
So please, try to clarify things first, instead of jumping to another thread.