1-pass Quality-based 75% Vs. 2-Pass Compressbility test 75% [Archive]

View Full Version : 1-pass Quality-based 75% Vs. 2-Pass Compressbility test 75%

lrk2

30th May 2007, 04:34

OK, this is my first post here, and yes I did extensive searching and found similar, but nothing directly answering my question...

Should I expect a better quality end result if I perform a 1-pass quality-based encode with 75% quality, or if I perform a 2-pass encode with a file size that gets a compressibility test value of 75% ? [all other things held constant].

My intuition tells me that I should get a better bang-for-the-bit using 2-pass encoding; However, I am looking for the best quality. I don't care that much (within reason) about the particular file size, but I want the best quality encode I can get (without being needlessly wasteful of disk space).

I have been using 2-pass, but to get the file size right with AGK, I am forced to pick a file size. I have been arbitrarily choose 2GB, and then aborting after the comptest, calculating the file size to get 75% comptest, and restarting. Am I wasting my time? I definitely get a smaller file size using this method than by using 1-pass quality-based 75%, but am I sacrificing quality? Maybe I'm delusional, but it seems to me that the encode quality looks better with 2-pass even with a smaller file size.

I'd really like to get the opinion of the experts on this forum.

Thanks!

nixo

30th May 2007, 06:59

Unless you're encoding for a standalone player, there's no reason for you to use 2 passes. Quality should be near identical.

*Edit* All settings being the same, file size should be near identical too - depending on how exact your comptest was.

--
Nikolaj

lrk2

31st May 2007, 19:51

Unless you're encoding for a standalone player, there's no reason for you to use 2 passes. Quality should be near identical.

That's interesting. I am not using a standalone player, but can you explain why 2-pass is required if I were to user one?

BigDid

31st May 2007, 21:00

... but can you explain why 2-pass is required if I were to user one?
Hi,

Answer is SIZE, with 2 pass you control the final size, with 1 pass you control quality only, not size.

Also if comparing same encode with same quality (ie 70%) with 2 passes you will have:
- less size
- a better quantizer distribution (more bits when needed less when not needed)

Did

nixo

1st June 2007, 02:41

That's interesting. I am not using a standalone player, but can you explain why 2-pass is required if I were to user one?

It is certainly not required but many standalone players are sensitive to bitrate spikes, and the only way that can be controlled with Xvid is using 2 passes.

--
Nikolaj

nixo

1st June 2007, 03:12

Hi,

Also if comparing same encode with same quality (ie 70%) with 2 passes you will have:
- less size
- a better quantizer distribution (more bits when needed less when not needed)

Well, I haven't done any tests but I think syskin is saying otherwise:
http://forum.doom9.org/showthread.php?p=817620#post817620

--
Nikolaj

lrk2

1st June 2007, 09:14

Well, I just did a test using Blood Diamond. The file sizes are very close to the same; However, to my surprise the 2-pass was actually ~57MB larger.

1-pass quality 75% : 2,352,582,656 bytes
2-pass comptest 75% : 2,412,025,856 bytes

The quality of the encode for both looks quite good to me.

Any way to find out the quantizer distribution?

nixo

1st June 2007, 10:35

loyalty_anchored

2nd June 2007, 04:09

interestingly enough i was about to post the same if not a very similar question, so instead of starting a new thread i will post in here. here is my current workflow for backing up all my DVDs:

1- use AutoGK in "target quality" mode at 75%
2- use the filesize of step 1 to do a "Custom Size" 2 pass encode

Am I wasting my time? I currently do not see a difference when playing back via my AppleTV on a 27 inch LCD TV, but I would like to be as prepared as possible for a future display upgrade.

Can someone with more experience then me tell me if I am wasting my time doing this 3 pass method?

Any explanation as to why my method is a good/bad one would be greatly appreciated.

I thank you all in advance for your help and knowledge.

EDIT: after re-reading the original posters message, i realised we are not talking about the same thing, no matter how similar it seemed originally. please accept my apologies.

lrk2

4th June 2007, 19:58

OK, I ran DivxDRFAnalyzer on both the 2-pass 75% comptest and 1-pass quality 75% encodes, I've pasted to reports for each below...but I think I need some help interpreting them. The biggest difference I noticed was the 2-pass has more frames falling into the DRF 1&2 category than the 1-pass.

1-pass has DRF=1&2: 17.4%
DRF=3: 34.7%

2-pass has DRF=1&2: 23.0%
DRF=3: 29.0%

Is one necessarily better than the other?

Here are the complete reports. There is also a graph that shows 20 second sections of DRF and bandwidth...but I'm not sure what to look for there either.

----------------------

1-Pass 75% Qualtity:

DivX DRF Analyzer v0.9.5 Report!
File Name: I:\Videos\Rips\BLOOD_DIAMOND_VTS_01_PGC1\BLOOD_DIAMOND.avi
FourCC: XVID
Codec: XviD0047
Resolution: [ Width: 704 Height: 288 ]
Frame Rate: 23.976 frames per second
The Video has 206184 frames [ 02:23:19 ]

Average Frame quality is MEDIUM [Average DRF/quantizer is 3.46]
Standard Deviation: Quality is MEDIUM [Std. Deviation is 0.96]
Image Resolution is HIGH

There are NO frame drops ( NO drops is better )

Recomended Resolution: [624x256] (Target DRF/quantizer=2.8)

Performance Caracteristics:
Macroblocks per frame: 792
The Width is multiple of 32

Kilobits per Second: 1691.70
Kilobits per Frame: 70.55
Kilobits per Macroblock: 0.089
Bits per Pixel: 0.36

Frame Type Statistics :
I Frames: 1.40%
P Frames: 51.31%
B Frames: 47.29%
S Frames: 0.00%
N Frames: 0.00%
(More Advanced Codecs use B and S frames)
Frame Quality Statistics :

DRF=1&2: 35329 17.4%
DRF=3: 70470 34.7%
DRF=4: 64347 31.7%
DRF=5: 33159 16.3%
DRF=6: 0 0.0%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.42%
KeyDRF<4: 2879
KeyDRF=4: 0
KeyDRF>4: 0

AverageKeyDRF: 2.67
MAXDRF: 5
AverageDRF: 3.46
Deviation: 0.96

----------------------

2-Pass 75% CompTest:

DivX DRF Analyzer v0.9.5 Report!
File Name: I:\Videos\BLOOD_DIAMOND\BLOOD_DIAMOND.avi
FourCC: XVID
Codec: XviD0047
Resolution: [ Width: 704 Height: 288 ]
Frame Rate: 23.976 frames per second
The Video has 206184 frames [ 02:23:19 ]

Average Frame quality is MEDIUM [Average DRF/quantizer is 3.40]
Standard Deviation: Quality is MEDIUM [Std. Deviation is 1.01]
Image Resolution is HIGH

There are NO frame drops ( NO drops is better )

Recomended Resolution: [624x256] (Target DRF/quantizer=2.8)

Performance Caracteristics:
Macroblocks per frame: 792
The Width is multiple of 32

Kilobits per Second: 1745.80
Kilobits per Frame: 72.80
Kilobits per Macroblock: 0.092
Bits per Pixel: 0.37

Frame Type Statistics :
I Frames: 1.39%
P Frames: 51.32%
B Frames: 47.28%
S Frames: 0.00%
N Frames: 0.00%
(More Advanced Codecs use B and S frames)
Frame Quality Statistics :

DRF=1&2: 46759 23.0%
DRF=3: 59057 29.0%
DRF=4: 65512 32.2%
DRF=5: 31983 15.7%
DRF=6: 0 0.0%
DRF=7: 0 0.0%
DRF=8: 0 0.0%
DRF=9: 0 0.0%
DRF>9: 0 0.0%
KeyF/DeltaF: 1.41%
KeyDRF<4: 2873
KeyDRF=4: 0
KeyDRF>4: 0

AverageKeyDRF: 2.65
MAXDRF: 5
AverageDRF: 3.40
Deviation: 1.01

lrk2

4th June 2007, 20:02

Oh, and the other thing I noticed from both reports was that the tool recommends dropping the resolution - I believe in order to decrease the average Quantizer from 3.4 to 2.8. What quality % would that equate to...and wouldn't that be overkill?

:thanks:

BigDid

5th June 2007, 20:57

Well, I haven't done any tests but I think syskin is saying otherwise:
http://forum.doom9.org/showthread.php?p=817620#post817620

--
Nikolaj
Hi,

Well yes (and I do respect Syskin a lot) and no.
Yes 2 pass encoding is not perfect and may not be able to achieve constant quantizer.
No What I am saying is IMO slightly different: a better quantizer distribution (more bits when needed less when not needed)

This was my experience when I did 1 or 2 tests encodes with first 70% and using the resulting filesize as a target for a 2pass encode: the quantizer where better distributed. I have verified that with DRF analyser and my eyes frame by frame.

This was also kind of academic cause I used and still use a Standalone player so 2 pass is needed :)

Did

BigDid

5th June 2007, 21:15

Oh, and the other thing I noticed from both reports was that the tool recommends dropping the resolution.. from 3.4 to 2.8.
Hi,

Drf analyser was made for older DivX rev so maybe at that time average Q 2.8 was needed.

If you browse the Xvid forum you will read that an Q 3 to 4 average is good to medium. If you need more quality, aim below Q 3.

Concerning your DRF comparisons, I see nothing wild. See my post above, for same filesize a 2 pass will vary the quantizer distribution (trying) to allocate more bits when needed (more Q1/2) and less bits when not needed (less Q3, more Q4 and above).

Now this is academic, you may want to compare some problematic frames like darkscenes, closeup, fast movement etc... frame by frame to see what suits you best.

Did

lrk2

6th June 2007, 03:45

Thanks for the info Did.

Yeah, I did a fast motion from comparison on Blood Diamond, but since the fast motion frames were very blurry to begin with, I couldn't really determine which was better or worse!

I just did an encode of a very dark movie - "The Good Shepherd" which I can compare. I've noticed a lot of artifacts and noise in dark scenes at 75% constant quality...but then the resulting bit rate is very low. The file size was 889MB for a 2hr 43min movie !

loyalty_anchored

6th June 2007, 17:53

Thanks for the info Did.
I just did an encode of a very dark movie - "The Good Shepherd" which I can compare. I've noticed a lot of artifacts and noise in dark scenes at 75% constant quality...but then the resulting bit rate is very low. The file size was 889MB for a 2hr 43min movie !

i noticed the same behavior when encoding "Panic Room" which is a very dark movie, resulting filesize at %75 was about 800MB, my average filesize on 40 encodes with AC3 audio is about 1.5GB so this one is pretty low.

surprisingly enough, it looks pretty good when played back on my display in the living room, watching it closely on my laptop screen, you can definitely see more artifacts then usual at 75%.

i am not sure if this is due to eyes blending dark artifacts to seem non-existant once the viewing distance is "normal", no one should be sitting nose to the screen anyways!

any comments?