DivX 5.1.1 Settings

[BiaTch 5.0]

I think it's time to revise whats good & what's not.

For a common 70-80% compressibility test or 2CD encode which settings are recommended

MPEG4 TOOLS:
B-Frames(?)
QPEL(?)
GMC(?)

Is PE fast(?)/slow(?) recommended yet?

Is setting bit-rate modulation to 0.02 still recommended (I was using it with 5.05)?

Thanks for any info.

[DigitAl56K]

B-frames recommended, QPEL is ok if you're happy to use only "Standard" p/q mode and don't want any support from DivX Certified devices (e.g. DivX DVD players), likewise with GMC - use it only if you're not worried about support for DivX Certified devices.

I couldn't recommend fixing the bitrate modulation to any particular value, the optimal setting is going to change based on the properties of the video you are encoding.

PV fast is useful if you are using very low bitrates but before you use it make sure (1) your video is not noisy and (2) your video is not interlaced (it should work if you de-interlace).

The guide (see the link in my signature) has a very good section on PV, I recommend giving it a quick read as it shows you how to evaluate PV for yourself.

[colordog]

Why do you say no PV on interlaced sources?

[DigitAl56K]

PV is not currently interlace-aware and so it might have a tendancy to degrade blocks where there is motion/combing effect assuming it to be texture erroneously.

[Voss]

I have been wondering about this question now that 5.1.1 is final and Xvid 1.0 are out.

I have been experimenting on Matrix Reloaded, going for a 2CD rip with the 5.1 AC3, 608x256 1083kb/s used for the last few tests. Comp test comes out at 62% IIRC.

I have tried DivX 3 times with:
-all standard settings
-with first pass encode speed standard, second slow (everything else std)
-and with QPel enabled on both passes.
The MV file was not used at all, and no PV effects either.

With DivX 5.1.1b1 and b2 the Qpel really helped make a cleaner image with more detail, and less speckles or ghosting around edges. To my surprise with 5.1.1 final the Qpel version was almost identical to the slow encode one.

The encoding times were of course very different for all 3 encodes, and I took numerous frames from each and compared back to back with ACDSee. In the slow scenes where the quant was on 2 they all look identical! I wasn't expecting that at all and it wasn't like that with 5.1.1b2. In the action scenes a slight difference was noticable between Qpel and Slow... Qpel had slightly softer image with a little detail loss (but only noticable on 2x zoom).

Can anyone confirm these tests... that the difference between the options is minimised? I doubt the movie is saturated at 62% and each enode was the same size, 1GB.

For those who care, encode speeds on a 3200+:
-standard; 1h10m roughly per pass
-second pass slow; just over 8h
-Qpel; almost 3h per pass

With my tests standard looks almost identical to the others and is many times faster.

[DigitAl56K]

I'll offer a few notes :)

When all three versions used Q=2 you wouldn't see any noticable difference because all three had processed the frame with the same "qualtiy". I say "quality" because this is what the quantizer most closely relates to.

QPel is nice, but its unsupported by DivX Certified devices. So, while you may think the QPel and 2nd-pass Slow encodes look identical in this case, only the 2nd-pass Slow version will actually play back on a DivX Certified DVD player (for example).

Slow (and Slowest) modes really offer the most benefit when you have particularly tight bitrate constraints. As your average bitrate rises they have a less significant overall effect.

Of course, you have to judge for yourself whether the increase in quality warrants using Slow or Slowest mode for any particular file/bitrate. Also try disabling post-processing entirely when evaluating your results - the PP algorithms might mask artifacts present in some of the encodings.

[Voss]

Thanks for that. I understand what a quantizer of 2 means, that's why I realise t+all 3 should look the same, but in the previous DivX vesion that wasn' tha case?

About DivX player support, honstly it doesn't concern me too much right now... some new players already support the extra features ;)

Ofcourse I used no PP while comparing, and actually made screenshots with VirtualDubMod to comperee the same frames.

To be honest all 3 looked identical while watching the actual movie. Only the screenshots showed some minor differances. I just wsn't expecting 1080kbs to be enough to equalise the results between the encodes. Overall the DivX quality is amazing, I am very happy with how far it's progressed.

[baz00ie]

Hello everyone, I just want to chime in here for all those who are like me and just require a simple straight forward method of making quality 2-pass encodes. i work with the freeware DivX5.1.1 - mainly for it's excellent combination of "ease of use" and "quality output".

I know that there are many here that have more experience than I when it comes to encoding and i also know that there are many who ask about "what's the best settings for.....etc.". I can offer neither. What i can offer is the "KISS" method - "Keep It Simple Stupid" - and from this method you'll be able to produce excellent quality encodes quickly and easily without much fuss.

(All that follows has been gathered from this forum, so thanks to all for your help)

Note about audio: I had decided early on to use only CBR MP3 - only for the reason that AVI does not officially support VBR. ( others have had success with VBR, but this is the KISS method :)

FIRST PASS
Once you have prepared you media with VirtualDub, AviSynth, or your application of choice, you are ready for your first pass throught the encoder.

In DivX5
set Performance/Quality to "Standard"
select "1-pass quality based"
set Quantizer to "2" or "97%"
activate "Write Log File"
de-activate "Psychovisual Enhancements"
all other settings are "default"
(bypass audio encode in your application)

* perform your first pass *

SECOND PASS
set Performance/Quality to "Standard"
select "Multipass, nth pass"
set encoding bitrate to 1448kbps
set Bitrate Modulation to "0.2"
select "Psychovisual Enhancements" if desired (Fast couldn't hurt)
all other settings "default"
Audio encode: 128kbps, CBR MP3

* perform second pass *

This should produce 1 hour of video per 700MB CD regardless of resolution.

(Note: This is NOT an automatic 2-pass method - you can't "set and forget". You must set-up and perform each pass seperately)


I have been using the method for almost a year now and i have been very happy with the results and the fact that it requires very little mental effort.

thanks for your time.
baz

[acrespo]

I never tried first pass with these settings. But I am curious. The quality based with quantizer 2 make a very big file (if your source file is divx too these intermadiate file can be larger than the source).
But I will do some tests to see differences between first pass "Multipass, 1st pass" mode and "1-pass quality based" mode.
The other settings are exactly of my videos.

[baz00ie]

Yes, the intermediate file will be large, but not too large - 1 hour usually makes about 1 - 2 gig file. Also, that file can be deleted after the first pass is complete - what we need is the log file that is created during the first pass.

cheers
baz

[Soulhunter]

Uhm... Is the resulting Quant2 log file not the same thing as making a regular 1stPass log file ???

Bye

[Voss]

That's what I was thinking also.... 1st pass does just that for you anyway, without making an actual 2GB avi file.

[DigitAl56K]

Originally posted by Soulhunter
Uhm... Is the resulting Quant2 log file not the same thing as making a regular 1stPass log file

Actually it isn't, afaik DivX will use a rate-control based encoding in "Multipass, 1st pass" instead of Q=2 (the XviD method). Also, you don't get any output of course in "Multipass, 1st pass", which is normally acceptable but consider live capture. You could capture from a live source using 1P-QB mode, writing a log file, then go on to recompress that file directly in "Multipass, nth pass", so its usefull to be able to run both methods :)

[Soulhunter]

Thanks for the info !!!

Thought DivX would do it the same way as XviD... ;)

Bye

[acrespo]

Originally posted by DigitAl56K
Actually it isn't, afaik DivX will use a rate-control based encoding in "Multipass, 1st pass" instead of Q=2 (the XviD method). Also, you don't get any output of course in "Multipass, 1st pass", which is normally acceptable but consider live capture. You could capture from a live source using 1P-QB mode, writing a log file, then go on to recompress that file directly in "Multipass, nth pass", so its usefull to be able to run both methods :)

I can use this method but I have a question. After capture I usually cut commercials and edit with avisynth noise filters (mipsmooth and bilinear resize). Is it possible in this case?

[Kalicrys]

Originally posted by acrespo
I can use this method but I have a question. After capture I usually cut commercials and edit with avisynth noise filters (mipsmooth and bilinear resize). Is it possible in this case?
Nope, sorry, the length and content of the scenes in the video will have changed, you have to do it after capture and avisynth.

Originally posted by DigitAl56K
Actually it isn't, afaik DivX will use a rate-control based encoding in "Multipass, 1st pass" instead of Q=2 (the XviD method).

Yup, they should be different. But after having the log filed modified to the target bit rate after the 2nd+ passes, what effect will this have on the quality of the resulting video compared to using a multipass 1st pass for the first pass in a multipass situation?

[theReal]

think it's time to revise whats good & what's not.

For a common 70-80% compressibility test or 2CD encode which settings are recommended

MPEG4 TOOLS:
B-Frames(?)
QPEL(?)
GMC(?)


I settled with these settings:
3 pass in "slow" mode (I think Gej from the Divx team said on their forum that even 2-pass in slow or slowest mode was enough for an almost perfect encode, so I think 3 passes should give you "perfect" Divx ;))

No QPEL, no GMC (I want to stay within the standards for hardware players)

psyvis never really helped me (especially not with 2CD encodes), so I don't use it neither.

I use pre-processing (usually "light", not more) for really noisy stuff like very old movies or tv captures.

[BiaTch 5.0]

Originally posted by theReal
I settled with these settings:
3 pass in "slow" mode (I think Gej from the Divx team said on their forum that even 2-pass in slow or slowest mode was enough for an almost perfect encode, so I think 3 passes should give you "perfect" Divx ;))

No QPEL, no GMC (I want to stay within the standards for hardware players)

psyvis never really helped me (especially not with 2CD encodes), so I don't use it neither.

I use pre-processing (usually "light", not more) for really noisy stuff like very old movies or tv captures.
Thank you for your input, I really want to get as much user opinions as possible

I have found pre-processing removes too much detail so I don't use it.

I'm now using 4 pass, 3 in standard (because standard goes so fast otherwise I would do 3 pass)& the last in slowest,

MPEG4 TOOLS:
B-Frames-YES
QPEL-NO
GMC-YES

& no PE.

I have to say GMC really seems to need a lot more CPU power to decode so I may stop using it.

[theReal]

Even the the slow mode needs so much more time to encode, I can't resist using it because it is using a new algorithm that is supposed to produce better quality than every version of divx before. Standard mode is more or less the same than Divx 5.05 in slowest mode.