AutoGk or Gk

[Panayotis]

Hi,

What is the difference between AutoGk and Gk?

Thanks a lot!

Stephane

NB: excuse me if this question has been already posted

[techz]

AutoGK is a one click solution for beginners, and GKNOT is a more complicated app with more options for the advanced user, i started with GKNOT and then AutoGK came out, u get used to gknot very quick if u read the great guides posted on the doom9.org site.

[JensG.]

You can get better quality with classic Gordian Knot mainly because you can encode main movie and credits separately. If you do not care for the last percent of quality you should be very content with AutoGK!

[gatormac]

Originally posted by JensG.
You can get better quality with classic Gordian Knot mainly because you can encode main movie and credits separately. If you do not care for the last percent of quality you should be very content with AutoGK!

Well, yes, but thats not the only reason. On most AutoGK encodes you can get quality comparable to a Gknot encode, but sometimes not having control over the codec can show up. I've tried AutoGK twice now....the first encode was Saving Private Ryan at 80% quality. It looks damn good and I'm very satisfied with it except for the filesize (5GB...is this movie impossible to compress?). I just did an AutoGK encode of The Matrix with the same settings and was disappointed. The dark areas looked too graduated and walls seemed to crawl/move as if they had life. I'm not sure, but it almost seems like Lumimasking, Qpel and GMC were enabled by AutoGK...which I never do as I don't like the effect they can have on a video. It also looked like a soft resize and soft matrix were used, but I know this isn't so as I checked the logfile and it shows that LanczosResize was used.

Anyway, there are other reasons to have more control, especially over codec settings.

[manono]

Hi-

...except for the filesize (5GB...is this movie impossible to compress?).

Probably one of the most difficult to compress movies, because of all the added grain.

Lumimasking, Qpel and GMC

AutoGK uses none of those. My guess is, that without using filters in GKnot to combat the things you describe, it would look pretty much the same way. Have you encoded that same movie in GKnot and compared the two?

[gatormac]

Originally posted by manono
Hi-

...except for the filesize (5GB...is this movie impossible to compress?).

Probably one of the most difficult to compress movies, because of all the added grain.

Lumimasking, Qpel and GMC

AutoGK uses none of those. My guess is, that without using filters in GKnot to combat the things you describe, it would look pretty much the same way. Have you encoded that same movie in GKnot and compared the two?

Actually, no I haven't, but I have looked at the AutoGK encode of The Matrix again and doggone it if it doesn't look fine! I really don't know what to say....I looked for the problems that I mentioned before and I just couldn't find them. I must be losing my mind. Has that ever happened to anyone before, seeing something wrong in the first viewing and then not being able to find anything wrong in the second viewing? Very strange.

Anyway, with that in mind I have 2 good encodes in 2 attempts. I'm glad to know that AutoGK doesn't use lumi, qpel, or gmc as I never do either. Which matrix does AutoGK default to when quality based encoding is used? Is trellis enabled?

The only thing i could see adding is simple noise reduction as I stated in another thread. Great program.

[piscator]

Originally posted by gatormac
Has that ever happened to anyone before, seeing something wrong in the first viewing and then not being able to find anything wrong in the second viewing? Very strange.

Yes, it has :D. You probably changed the decoder or some decoder settings between the two views.

greetz,
Piscator

[manono]

Hi-

Which matrix does AutoGK default to when quality based encoding is used? Is trellis enabled?

No Trellis. For Quality based encoding (unless you do the special install for people with ESS chipset standalones), it's HVS_Best matrix. For 2-pass Predefined size encodes, when the results of the Compress test are low, it'll switch to HVS_Better.

[gatormac]

@manono

Thanks for the answers....it is nice to know what settings are being used. I have another for you (or anybody who knows):

When using the Gknot/VDub method I would run a compression test at a certain resolution, then set the comp% to 80% and encode. Is this how the Quality based encoding is done (using AutoGK I chose 80 for quality) or is 80 just a setting for a constant quantizer (I think the tutorial states that 67% is quantizer 3). If the latter, what would be the quantizer for 70,75,80%?

[manono]

Hi-

...then set the comp% to 80% and encode.

How are you getting to 80%? Are you lowering the resolution, or increasing the file size? The GKnot method you describe isn't equivalent to the Quality based encoding of AutoGK. GKnot uses 2-pass methods to achieve a given file size. Auto GK uses the formula as described in Question 13 of the XviD FAQ (http://forum.doom9.org/showthread.php?s=&threadid=16935):

((MaxQuant-MinQuant)/100 * (100-quality)) + MinQuant

What this means to you is that when using Quality based encoding in AutoGK, that it'll switch back and forth between quants 2 and 3, with the percentage of each determined by that formula. In your 80% case, the final average quant will be 2.2, which is very high quality.

I think the tutorial states that 67% is quantizer 3.

It does say that, but I'm no longer sure that's correct. That reminds me that I have to check with len0x about that.

[gatormac]

@manono

To get to 80% I was increasing the filesize. I saw that equation in the FAQ, but I didn't know how to apply it....if I want a 70% encode, what are the Min and Max Q's? Anyway, its nice to know how it all works. When not using AutoGK I have been toying around with constant quantizers between 2 and 3. I just did an encode of LOTR:FOTR at a Q of 2.5, maintaining the AC3 and only cropping the source maintaining the original resolution, encoded with a PAR of 16:9 NTSC. Final filesize 2.6GB, plays perfect in MPC and WMP.

Thanks for the explanations!

[Gordoran]

two little questions:

1. less quantizer is better?
2. 7 0 178 2 2 2 2 2 2 2 2 2 2 2 2 <--- do i see here the quants ?

[manono]

Hi-

I didn't answer the previous question because I didn't know the answer. I don't know what quants AutoGK uses for max and min all the time, or if it varies depending on the percentage chosen. But I know the answers to your new questions.
1. Yes, the lower the better. For all practical purposes, 2 is the lowest and best. There is a quant 1, but the size that it produces is so large, that it isn't useful.
2. No, that's from the .d2v, and the numbers in there have nothing to do with quant values. To learn what those numbers mean, please see Question 8 of the DVD2AVI FAQ. (http://forum.doom9.org/showthread.php?s=&threadid=59272)

[Gordoran]

ah kewl,
but where i can see the quants i use for my movie?

[r6d2]

Originally posted by manono
It does say that, but I'm no longer sure that's correct. That reminds me that I have to check with len0x about that.It was the last time I checked.
Originally posted by Gordoran
but where i can see the quants i use for my movie?You cannot see the quants you use, unless you know beforehand len0x's formula which is Quant = 200/quality (only valid in quality based encoding).

But you can always see the quants you got with ffdshow. Very illustrative. Shows optionally quants and motion vectors on a macroblock basis while the movie is played.

[Gordoran]

aehm thx, but now im really confused :rolleyes: :D

[manono]

Another tool you might find useful to help dissect your movie is DRF Analyzer (http://www.geocities.com/analyzerDRF/). DRF (Detail Removal Factor) is another name for Quantizer (or Quant, for short).

[Gordoran]

ah nice prog, thx.

if my DRF shows a lot of yellow or red, its better to increase filesize or decrease resolution right?

[manono]

Hi-

I'll assume that you're using B-Frames. I find the program somewhat misleading where B-Frames are concerned. It shows the B-Frame quants themselves, rather than the quants on which the B-Frames are based. In any event, it's easy enough to figure out which of those yellow and red colors are the B-Frames (all or almost all of the red ones certainly, and some of the yellows), and to ignore them. Another way to get the quant values , and for the B-Frames, the quant values of the P-Frames on which they're based, is to run DebugView (http://www.sysinternals.com/ntw2k/freeware/debugview.shtml). That has to be done while the encoding is going on, though.

AutoGK limits the quants to 2/6 2/6 (I think), and the B-Frame settings are 1/150/100/0 (default). They're in the B-Frame Control portion of the Global Tab. Hold your mouse over the respective numbers to see what they mean.

You can't do anything about the AutoGK settings (which are pretty good all-purpose values), but if you're using GKnot, or encoding directly in VDubMod, you can get rid of all the red, and a lot of the yellow (unless you're giving the end credits high quants), by doing such things as restricting the quant spread severely, using even more conservative B-Frame settings, and/or getting rid of B-Frames entirely. I don't necessarily recommend doing any of those things. But if you're interested in improving the look of that graph in DRF Analyzer, the first thing to do would be to change the quant spread from the default 2/31 2/31 (if that's what you use). You could just redo the second pass of a movie you've already done, and use something like 2/3 2/4. If the compress test gave you a decent percentage (over 50% or so), then you won't wind up oversized.

[Gordoran]

aehm, my english oh my english :)
I have to read several times to understand, but didnt get it all :D

I use GKnot with XVid and set both passes with same settings, bad or good?
Thats all I need to know, my rips are good looking atm, anything better would be kewl but not very necessary :)

[manono]

Well, I can't really tell you how to encode in GKnot. I don't use GKnot for encoding myself, but encode directly in VDubMod. And my methods are unorthodox at best, and heresy at worst. But my methods are based on 2 beliefs; that XviD's rate control is broken, and that quality based encoding gives you the best and most even results, and that the best way to approximate quality based encoding when using 2-pass methods is by limiting the quants.

So, the first pass settings don't need changing. But if you want to try what I suggested in the previous post, when setting up for the second pass, go into the Quantization Tab, and change the quantizer restrictions (from top to bottom) to 2,3,2,4. All this is with Koepi's 6/24/03 version of XviD (the one included with AutoGK), and not the 1.0 RC versions. By using those settings, you'll get no red frames when looking at the graph in DRF Analyzer (except maybe for end credits, if they're encoded with high quants).

[len0x]

Originally posted by manono

AutoGK limits the quants to 2/6 2/6 (I think),

Not really :)
it's 2/3 2/3 actually (if first pass is more 66% or 2/4 2/4 if less then that)

P.S. this all reminds me to check if I should relax this for new XviD 1.0 or not... Have you started using new XviD yet, Manono ?

[manono]

Hi len0x-

OK, thanks for the info. That's good to know. And no, I'm still sticking with old reliable Koepi's 6/24.

[Gordoran]

Originally posted by manono


So, the first pass settings don't need changing. But if you want to try what I suggested in the previous post, when setting up for the second pass, go into the Quantization Tab, and change the quantizer restrictions (from top to bottom) to 2,3,2,4. All this is with Koepi's 6/24/03 version of XviD (the one included with AutoGK), and not the 1.0 RC versions. By using those settings, you'll get no red frames when looking at the graph in DRF Analyzer (except maybe for end credits, if they're encoded with high quants).

ah, got it, but why only in the second pass and not in both? :)

[manono]

Because the first pass is always encoded at quant 2, and in the Quantization Tab the quantizer restrictions are all greyed out. Something you might want to try is to save the first pass as a playable .avi. When setting up the first pass, go into the Two Pass Tab, and uncheck "Discard first pass". If you open that in DRF Analyzer when done, you'll see the I and P Frames are all quant 2, and the B-Frames are higher. How high depends on the B-Frame settings, but with the default settings, they'll all be quant 4 (I think).

[Gordoran]

ah, ic, now im glad :)


/€
hmm, over here they are not greyed out :confused: