Difference: 2nd pass of 2pass vs 2nd pass of 3pass!

[kotuwa]

What is the difference between the 2nd pass of the 2 pass and the 2nd pass of the 3 pass encode....

I did some samples and there, both are same if --slow-firstpass used in the first pass.

But if the --slow-firstpass is not used, then the output files are different.

Need to know what are the internal differences.


Idea is to do a 2-pass with stat files saved.
But in any case if I would want to do another pass, it is really good if it is a 3 pass... And to do that, using 3 pass from the beginning even though initial plan is to go with only 2 passes...

What lacks in the 2nd pas of 3 pass compared to the 2nd pass of 2 pass?
(It would be the answer to why having 3 options as 1, 2 and 3?... If it is same, no need of a 3rd one... But I need to know..)

Thank you very much!

[poisondeathray]

The difference is --pass 2 does not overwrite the stats file

In all cases, --pass 2 signals the "last" pass. So in the 2pass encode it's the final pass, and the stats file is not overwritten. In the 3 pass encode it's the --pass 2 is the actually the 3rd pass. The numerical 2nd pass is --pass 3 in the 3pass encode. For example, in a hypothetical 10pass encode, the 2nd to 9th passes are all --pass 3

ie. the numerical 2nd pass for the 2pass encode doesn't have the stats file overwritten for the 2nd pass, but the 3pass does. This means you cannot just add additional "n" passes for a 2pass encode that had already used --pass 2 - because as soon as the encoder "sees" --pass 2 command, that pass is no longer checked for convergence - it's used as the final pass (it's not equivalent to the 2nd pass of a 3 or more pass encode)

[LigH.de]

A warning regarding 3 (and more) passes: It will refine the bitrate distribution, but don't be certain that the results will be optimized towards a convenient result for your subjective quality impression. Not everyone will agree that "more passes" will also mean "better quality retention"; like "preset placebo" it may just be a waste of time.

[kotuwa]

poisondeathray Got it,
but still not clear why the output of 2nd pass that uses --pass 2 is better than the output of 2nd pass that uses --pass 3 !!!
I think in other words, the advantage of telling that this is the final pass...

Also asking that if so, then using --pass 2 in the last pass of a 3pass encode is better than using --pass 3, right? (there the difference could be minimal even compared to the above main situation where it is very small too)

A warning regarding 3 (and more) passes: It will refine the bitrate distribution, but don't be certain that the results will be optimized towards a convenient result for your subjective quality impression.

I think you got this a wrong way....
Are you telling that doing a 2-pass twice is generally better than doing a 3 pass once? (no time advantage is there, same time... in first, doing 2pass twice, the 1st 2pass is completely wasted, in 3 pass, it updates the stat file at least)
Because the question about this scenario! (last part of the OP)

[LigH.de]

A 2-pass twice? No, that would be nonsense.

I mean that a 3-pass encode does not necessarily "look better than" a 2-pass encode. Even though the last pass will work with an additionally refined statistics file.

Different people may judge differently, and it may also depend on the material and limits (bitrate, complexity) if you will notice a difference at all, and whether you would subjectively prefer the 3-pass result over the 2-pass result.

[poisondeathray]

poisondeathray Got it,
but still not clear why the output of 2nd pass that uses --pass 2 is better than the output of 2nd pass that uses --pass 3 !!!
I think in other words, the advantage of telling that this is the final pass...

Also asking that if so, then using --pass 2 in the last pass of a 3pass encode is better than using --pass 3, right? (there the difference could be minimal even compared to the above main situation where it is very small too)

You already answered it yourself, just stop and think about it:

--pass 3 uses the "fast" analysis pass settings, not the user selected encode settings (unless --slow-firstpass is used)

--ref 1
--no-8x8dct
--partitions i4x4 (if originally enabled, else none)
--me dia
--subme MIN( 2, subme )
--trellis 0

[kotuwa]

You already answered it yourself, just stop and think about it:

--pass 3 uses the "fast" analysis pass settings, not the user selected encode settings (unless --slow-firstpass is used)

--ref 1
--no-8x8dct
--partitions i4x4 (if originally enabled, else none)
--me dia
--subme MIN( 2, subme )
--trellis 0

No, it did not!!...
In my tests, the mediainfo encoding settings are same in --pass 3....
It is not act like --pass 1 as you have mentioned... no turbo-like settings...
Are you sure?
Is there any builds behave like that?...
And also MeGUI sends the command --pass 3 always, even for the final pass (3rd pass of the 3pass)....
If you are right, then its outputs would be useless

[poisondeathray]

EDIT: ok you might be right about --pass 3, not using "fast" settings. Let me double check

Yes, you were correct --pass 3 isn't a "fast" pass like --pass 1 by default





Post your megui log file of an example that shows this behaviour of --pass 3 for the final pass

[poisondeathray]

They are the same. Same bitrate and filesize. Same frametypes. Same frame sizes. Same decoded image. Same per-frame PSNR . I tried it on several samples, a few short ones then a few longer ones in case a longer "horizon" might influence the frametype distribution. All the same

"normal" 2pass encode vs. "interrupted" 2nd pass of a 3pass encode (--pass 3 for 2nd pass)

Can you post some more information about what you think is different, or how you came about that observation, or source samples that caused a difference

[Detmek]

Frametype decision is done during first pass so --pass 3 will not change that.

[kotuwa]

@poisondeathray difference mean output file size.....

2nd pass --pass 2 - 1.77 MB (1,856,713 bytes)
2nd pass --pass 3 - 1.76 MB (1,855,590 bytes)

If you mean same, did you check sizes?
If same, in any case have you used --slow-firstpass in the 1st ass?
If so, then it gives exactly identical outputs... (like I said in OP)

[poisondeathray]

Yes, I mentioned filesize identical. Even PSNR per frame identical . All four test samples

This is with default settings (not slow first pass)


Here is one example

2nd pass --pass 2 25.5 MB (26,799,381 bytes)
2nd pass --pass 3 25.5 MB (26,799,381 bytes)

("normal" 2 pass encode)
2nd pass --pass 2
x264 [info]: frame I:48 Avg QP:22.22 size: 27381
x264 [info]: frame P:2755 Avg QP:25.23 size: 7413
x264 [info]: frame B:2321 Avg QP:27.78 size: 2155
x264 [info]: consecutive B-frames: 18.3% 58.2% 16.8% 6.6%
x264 [info]: mb I I16..4: 6.6% 78.5% 14.9%
x264 [info]: mb P I16..4: 1.5% 6.7% 0.9% P16..4: 49.3% 16.1% 7.4% 0.0% 0
.0% skip:18.1%
x264 [info]: mb B I16..4: 0.1% 0.3% 0.0% B16..8: 47.5% 4.7% 0.8% direct:
1.4% skip:45.2% L0:37.1% L1:54.4% BI: 8.5%
x264 [info]: 8x8 transform intra:74.6% inter:81.1%
x264 [info]: coded y,uvDC,uvAC intra: 68.8% 69.3% 25.0% inter: 21.4% 22.0% 0.6%
x264 [info]: i16 v,h,dc,p: 23% 33% 5% 39%
x264 [info]: i8 v,h,dc,ddl,ddr,vr,hd,vl,hu: 22% 13% 23% 5% 7% 9% 7% 8% 7%
x264 [info]: i4 v,h,dc,ddl,ddr,vr,hd,vl,hu: 24% 25% 16% 5% 7% 8% 6% 6% 4%
x264 [info]: i8c dc,h,v,p: 52% 19% 23% 6%
x264 [info]: Weighted P-Frames: Y:3.8% UV:1.5%
x264 [info]: ref P L0: 65.8% 19.1% 10.9% 4.0% 0.2%
x264 [info]: ref B L0: 90.4% 9.2% 0.4%
x264 [info]: ref B L1: 97.6% 2.4%
x264 [info]: kb/s:1000.91

("interrupted" 2nd pass of 3pass encode)
2nd pass --pass 3
x264 [info]: frame I:48 Avg QP:22.22 size: 27381
x264 [info]: frame P:2755 Avg QP:25.23 size: 7413
x264 [info]: frame B:2321 Avg QP:27.78 size: 2155
x264 [info]: consecutive B-frames: 18.3% 58.2% 16.8% 6.6%
x264 [info]: mb I I16..4: 6.6% 78.5% 14.9%
x264 [info]: mb P I16..4: 1.5% 6.7% 0.9% P16..4: 49.3% 16.1% 7.4% 0.0% 0
.0% skip:18.1%
x264 [info]: mb B I16..4: 0.1% 0.3% 0.0% B16..8: 47.5% 4.7% 0.8% direct:
1.4% skip:45.2% L0:37.1% L1:54.4% BI: 8.5%
x264 [info]: 8x8 transform intra:74.6% inter:81.1%
x264 [info]: coded y,uvDC,uvAC intra: 68.8% 69.3% 25.0% inter: 21.4% 22.0% 0.6%
x264 [info]: i16 v,h,dc,p: 23% 33% 5% 39%
x264 [info]: i8 v,h,dc,ddl,ddr,vr,hd,vl,hu: 22% 13% 23% 5% 7% 9% 7% 8% 7%
x264 [info]: i4 v,h,dc,ddl,ddr,vr,hd,vl,hu: 24% 25% 16% 5% 7% 8% 6% 6% 4%
x264 [info]: i8c dc,h,v,p: 52% 19% 23% 6%
x264 [info]: Weighted P-Frames: Y:3.8% UV:1.5%
x264 [info]: ref P L0: 65.8% 19.1% 10.9% 4.0% 0.2%
x264 [info]: ref B L0: 90.4% 9.2% 0.4%
x264 [info]: ref B L1: 97.6% 2.4%
x264 [info]: kb/s:1000.91


as expected, the continued "normal" 3rd pass with --pass 2, refines the bitrate target and it's closer to the bitrate target of 1000kbps
actual 3pass
x264 [info]: frame I:48 Avg QP:22.17 size: 27667
x264 [info]: frame P:2755 Avg QP:25.25 size: 7404
x264 [info]: frame B:2321 Avg QP:27.78 size: 2155
x264 [info]: consecutive B-frames: 18.3% 58.2% 16.8% 6.6%
x264 [info]: mb I I16..4: 6.4% 78.8% 14.8%
x264 [info]: mb P I16..4: 1.5% 6.8% 0.9% P16..4: 49.1% 16.1%
.0% skip:18.2%
x264 [info]: mb B I16..4: 0.1% 0.3% 0.0% B16..8: 47.4% 4.7%
1.4% skip:45.2% L0:37.1% L1:54.4% BI: 8.5%
x264 [info]: 8x8 transform intra:74.5% inter:81.0%
x264 [info]: coded y,uvDC,uvAC intra: 68.8% 69.4% 24.9% inter: 21.4
x264 [info]: i16 v,h,dc,p: 23% 33% 5% 38%
x264 [info]: i8 v,h,dc,ddl,ddr,vr,hd,vl,hu: 22% 13% 23% 5% 7% 9%
x264 [info]: i4 v,h,dc,ddl,ddr,vr,hd,vl,hu: 24% 26% 16% 5% 7% 8%
x264 [info]: i8c dc,h,v,p: 52% 19% 23% 6%
x264 [info]: Weighted P-Frames: Y:3.8% UV:1.5%
x264 [info]: ref P L0: 66.9% 18.4% 10.5% 4.0% 0.1%
x264 [info]: ref B L0: 90.2% 9.4% 0.4%
x264 [info]: ref B L1: 97.6% 2.4%
x264 [info]: kb/s:1000.54

[kotuwa]

what is the build?
and what does >"interrupted" 2nd pass of 3pass encode< mean? (me noob)
!?

[poisondeathray]

I used komisar's build, but slightly older version r2525
http://komisar.gin.by/

"interrupted" means you don't finish the 3rd pass . You stop after the 2nd pass using --pass 3 , before you begin the actual pass 3.

Because that is the topic that you started isn't it? Comparing (final, normal) 2nd pass of using --pass 2 , to the 2nd pass of 3pass using --pass 3 . If you don't "interrupt" it, you will be using pass 3, not pass 2.

[LigH.de]

"Interrupt" may not be a suitable term here, it suggests that you interrupted (Ctrl+Break) a running middle pass so it didn't finish; but you mean that you finish the middle pass but omit the final pass.

[kotuwa]

yeah... but why interrupting... it is like stopping in the midle of something....
I think you consider a pass 3, as one unit, all 3 passes as 1 unit... while it can be considered as 3 separate parts :]
What you did, is it automated or something?
because like LigH.de said, >"interrupted" 2nd pass< looks like you closed the program or ctrl+break in the middle of the 2nd pass!
anyway, it is not that important regarding the results
:]

[poisondeathray]

Normally, when you do a 3pass or more encode, you do it with a batch file

--pass 1
--pass 3
--pass 3
.
.
.
--pass 2

You don't do 1 pass at a time, then come back to the computer, manually run the next pass, etc...You do it all at once.

That's why it's "interrupted" . It's placed in quotation marks to signify the difference. If I didn't "interrupt" the 3pass encode, I'd be looking at the 3d pass, not the 2nd pass with --pass 3. In the context of this thread topic, I thought it was pretty clear. I guess not...sorry for the confusion





Did you get something different in the 2nd pass ?