Option "Full Quality first Pass"?

[guada 2]

Hello to all,

It seems to me that it is possible to notch this option, but in which particular case should I use it truly to enjoy his potential?

Which are then the adjustments which you could propose to me?

Thank you in advance.

[guada 2]

I found this like explanation (Source: CRUSTY FAQ)

- ' Full quality first pass':By default unquestionable XviD switches options off in the first pass to speed it up has bit.The options that are switched off are not really necessary for has ' normal' first pass (normal ace in:you don' T keep it afterwards) and turning them off edge increase encoding speed considerably.
However this is has relatively new option and some people are concerned that the second pass might cuts has off little bit less quality because of the options turned in the first pass.Since the jury is still out one this one you edge set this option to make the codec uses all the options for both master keys, yew it makes you feel comfortable.Recommended setting is off so the codec edge speed things up has bit.

However, I have a question:
Which interest to keep this option in the configuration of the codec Xvid?

[Mnl]

If you disable full quality first pass the file you get after first pass is just a dummy file containing the information xvid needs to make the second pass - that is to say, you cannot play the file in a media player.

If you enable full quality first pass the file you get after first pass is a normal video file made at quant 2 with all the options you enabled for xvid.

That is to say if the filesize is small enough after first pass and you enabled full quality first pass you don't have to run the second pass - just use the first pass file.

At least I think that is what the option does.

[tedgo]

If you want to get a working file at first pass by enabling "Full quality first pass", you have to disable "Discard first pass" too.

[guada 2]

I thank you both for your point of view.

Indeed, I make this action well, but it does not bring anything serious to me.
It is for that that I wanted to know how to fully benefit from this option like it belongs to the configuration of the codec.
Therefore, I wondered then if it does not make distinction obvious remove it as much.

[Brother John]

Mnl, not exactly. What you described is the "discard 1st pass" option. When "discard" is disabled you always get a playable video. But if "Full quality" is disabled it might not be MPEG-4 compliant.

So "Full quality" is primarily useful, if - for some reason - you want to keep the 1st pass.
Apart from that i'd always leave it unchecked because of the very noticeable speedup. And regarding quality: The feature has existed long enough to rule out severe impact on quality. Certainly someone would have found such problems by now.

[Koepi]

I for one use full quality first pass because I can see a difference if I compare side-by-side encodes with and without that. It's just a small improvement though - so after all it's a matter of taste. If you "waste" hours over hours for getting a perfect encode, you should use a full quality first pass, too. If you just want a fast, good quality encode, switch it off.

Cheers
Koepi

[Mnl]

Hi Brother John

Thanks for the clarification. I had a feeling that my answer wasn't entirely correct :o. Must remember not to post when I'm not 100% sure :)

[guada 2]

Thank you Koepi,

Which pleasure to see you again.
I must say to you that I have do much test of Xvid and yet nothing enables me to note a quality which slices.
It should be said that my tests are articulated on low bitrates.
It would undoubtedly be necessary that I exploit the upscaling with IIP of Didée with this option.
Nevertheless the last version: Xvid CLI is not very stable but increases quality.

A small report:
Method: First Pass with the Stable Version and a Second Pass with the Last Version is better than 2 Pass of the last version and lower size.

Of course, I tested all the confused modes, it is that which makes the difference.

[IgorC]

I was playing with FullQ1pass and some SSIM tests too. FullQ1pass even gives a bit lower SSIM (- 0.02). Visually there is no difference.
However its decision of everyone.

[Koepi]

Quality metrics like PSNR and SSIM don't tell the whole story.

I can't stress the example enough that we had with the x_THRESHOLD_TOO_LOW=3 -thingy. PSNR went up quite considerable, and the image was all blurred. It was lousy. (Also, a change in the SSIM on the second after-comma digit is within the precision of measurement.)

But suite yourself and believe in the magic numbers ;-)

Cheers
Koepi

[Didée]

I'll try a short explanation.

During 2nd pass, the quantizer for each frame is chosen based on how big the frame was during the 1st-pass encoding.

The point is, that due to make the 1st pass perform fast, many advanced features of XviD are disabled. This means that for each frame, the compressability in the 1st pass is different from the compressability in the 2nd pass (since 2nd pass has more tools available to achieve a "better" compression).
Therefore, the internal calculation of "how big a frame will come out when compressed with quant=X during 2nd pass" is less precise with a normal "fast" 1st-pass, and more precise when using "full quality 1st pass".

Most times, the difference is neglectable. The calculation how big a frame will become is never perfect, but a well educated guess. To deal with these (mostly rather small) differences between "calculated" and "effectively achieved" frame sizes, that's the job of XviD's Overflow Treatment routines.

In the end, overflow treatment has a little more work to do when a fast 1st-pass was done, and a little less with a full-Q 1st pass. This means that the distribution of quantizers is less optimal with a fast 1st-pass, and better with a full-Q one.

However the difference is really not erthbreaking, but rather sophisticated instead. We're mostly talking about quantizers for very few frames in the whole clip that would change by +/- 1, mostly. For P-Frames, that is.

[Sagittaire]

IMO why use full first pass:

- perhabs better bframe number decision
- perhabs better GMC decision
- better quant prediction

but it's very small improvement and visually it's impossible to see the difference if you use fast first and full first pass ... full first pass doesn't change ME or RDO in second pass for example ... it's just a little RC tweak ... lol

If difference between first pass average quant and second pass average quant is high, fast first pass with higher quant will be a very better RC tweak than full first pass in q2 ...


I can't stress the example enough that we had with the x_THRESHOLD_TOO_LOW=3 -thingy. PSNR went up quite considerable, and the image was all blurred. It was lousy. (Also, a change in the SSIM on the second after-comma digit is within the precision of measurement.)


"considerable metric improvement" ... with OPSNR and SSIM ... lol ... I'am very curious to see that ... try with my HPII trailer ... pleeeeeeeaaaaaaase ... ;)


and less blur does not mean always better quality

more blur, better metric ... and better visual quality
http://multimediacom.free.fr/Test/ffdshow/PostOn.jpg

less blur, worst metric ... and more block
http://multimediacom.free.fr/Test/ffdshow/Postoff.jpg

[guada 2]

Thank you Didée, Sagittaire, Koepi and IgorC,

@ Sagittaire

If a difference exists it would not be not visible with low resolution: useful for a QP and a beneficial "GMC" on scenes with displacement horizontal...

A test on a "ANIME" with the option "Cartoon activated mode" makes it possible it to visualize a difference or not?

Other question:
The "Turbo" option is it really useful in the analysis of the movements when options B frame and QP are activated?