compressability checks

[SiXXGuNNZ]

ok, I know to search, I did, I also read the guides, I am more concerned with the %

guide says 50-60 is good, while 80 is overkill.

so, would 50-75 be a good range?

any help is appreciated :)

[Hiro2k]

Yes it's a good range. I try to go for 60-75% for 2 CD, if it's higher, I go for 1 CD.

[SiXXGuNNZ]

Originally posted by Hiro2k
Yes it's a good range. I try to go for 60-75% for 2 CD, if it's higher, I go for 1 CD.

thanks for the answer, I never used compress check till today, never understood what it was for, and after reading some things on it, I felt that I was missing something.

wouldn't 60-75% percent be good for 1 and 2 cd rips, if not, are you saying a 1 cd should always be above 75?

[SiXXGuNNZ]

actually I think I now understand what the % represents. it is how close it will look to the original source, so, technically, 100% would be good if it was 1 cd right? obviously that would rarely if never happen. since resolution is always important. on a more important note, all my source material is NTSC, so 720x480, for 16x9 and 4:3, I always try to stay as close to 100% on the resize as I can. what is a general area that is decent to drop this to? I normally work with 704 by xxx on my 16x9 sources. I have seen ppl say they like to work with 640 by xxx, how low can one really go before the movie just becomes too small to watch fullscreen?

I am currently working on "The Untouchables", for a 1 cd rip I can get:

93.5% at 544 by xxx (0.237 bits/pixel)
82.4% at 576 by xxx (0.209 bits/pixel)
73.2% at 608 by xxx (0.185 bits/pixel)
65.4% at 640 by xxx (0.166 bits/pixel)
58.8% at 672 by xxx (0.149 bits/pixel)
56.2% at 704 by xxx (0.142 bits/pixel)

now, I am safe to assume that going with what the guide reccomends of 0.220 bits/pixels I can safely choose the 93.5%?

or should I also listen to the guide and not go over 80%

this is where I truely get confused I suppose.

[SiXXGuNNZ]

see I do not understand, the compress check the avg bpp was 0.253 at 704 by xxx and 56.2%

I really do not understand at all :(

[N_F]

I don't understand your last post, still some random thoughts for you:

The absolut minimum for non-fullscreens I may use is 512W, but that's only in extreme situations. Normally I avoid going lower than 576W. For fullscreens the situation is a bit different.

If I had those percentages you show I'd either go for 608W or 640W, depending on which gave me the smallest AR error.

Did you make a new comp. test for each resolution or did you just move the slider? If you just moved the slider you have the wrong numbers.

A comp. test doesn't show how close to the original source you are. Rather how close you are to a quantiziser 2 picture.

A very high comp. value with a very low resolution is no good. A very low comp. value with a very high resolution is no good either. It's you're job to find a good middle.

[SiXXGuNNZ]

no, you were correct, I just moved the slider

you said:

A very high comp. value with a very low resolution is no good. A very low comp. value with a very high resolution is no good either. It's you're job to find a good middle.

so with going with the guide, it says 50-60%, anything over 80% is not good. so am I safe to assume that Hiro2k using 60-75% is a good value?

or should I always stick with 50-60%

I am sorry if I am being annoying, I am really having a hard time understanding comp tests :\

[Hiro2k]

<sarcasm>
It's kinda fun for us. Cause we all love so see all newb's struggle
</sarcasm> :)

Ok so first of all, you have to decide what percent is suitable for you. Then you can base all your decsions from there. What I said was that if you have a 2CD movie with a % above 85% you should make it 1CD with a lower %, but your save space at the cost of lowering the resolution a bit.

[SiXXGuNNZ]

Originally posted by Hiro2k
<sarcasm>
It's kinda fun for us. Cause we all love so see all newb's struggle
</sarcasm> :)

Ok so first of all, you have to decide what percent is suitable for you. Then you can base all your decsions from there. What I said was that if you have a 2CD movie with a % above 85% you should make it 1CD with a lower %, but your save space at the cost of lowering the resolution a bit.

hehe, sarcasm is ok :)

I am a newbie, altho I have been encoding with mpeg4 for 6 months(yet I still know I am not a guru yet :D), like I said I always just used the bpp thing at .220 for 1 cd rips never using the comp test :)

I am just having alot of trouble grasping what comp tests really mean I guess, and from doing my forum rules searching, I found alot of threads where others also have trouble grasping what it is that it represents. and the guides really do not explain what it actually means, just some guy saying between 50-60%, and yes I grasp that by posting this I am in a way asking one or two guys for their opinions on it. I guess I was trying to actually understand what the comp test really means, but if you guys say between 50-60% I guess between 50-60% it is.

maybe there is a page that can help me understand them better? like, the tooltip for %, almost exact quality indicator, what does that actually mean? I also see it gives a bpp for the test, currently my res says 0.209, the test says, 0.290, so should I be actually trying to get the comp check test bpp down to the guides reccomended 0.22?

I re ran my comp tests, here are the results.

576x240 = 72.0% of 0.290
608x256 = 66.9% of 0.277
640x272 = 62.3% of 0.266
672x288 = 58.3% of 0.256
704x288 = 56.2% of 0.253

edit: I messed up and forgot to do to comp checks :\ doing them now

so which one looks like the winner? and why?

I really do appreciate you guys taking time to deal with the newbs :D

[folio]

after youve done the compress test ignore the Bit/pixels value, look at the % value the test brought back to you....for 1 cd stick to between 60 and 70 % anyhting above might not be that good for 1 cd ...in your last test for example the resolution of 608x256 would be a good starting point...only use the bits/pixel frame as starting point before doing the compress test...i.e. use the slider until you get the value bwtween 17 and 20 (1 cd) and then use compress test and go from there!
good luck!

p.s. everyones a newb at this cos you never know whats around the corner!!

[SiXXGuNNZ]

thank you to all the ppl who gave me feedback, I will do an encode at 608 by xxx, I appreciate the help :)

[Acaila]

Let me take a shot at trying to explain exactly what a compressibility test does.

When you start a Compressibility Test (CT from now on), GKnot will order the codec to take parts of the video and encode them at the maximum quality (if you look closely at the Avisynth script generated by GKnot you'll see it adds SelectRangeEvery(280,14) to the script, which means it will take 14 consequetive frames at intervals of 280 frames, that's what is approximately 5%). The length of the video that you get after the test will be extrapolated to a full-size video (so for 5% CT that's current_size*20). This will be your video at 100%. Now, when you select a target size for your video on the bitrate tab GKnot will calculate target_size/full_size, this value is your CT percentage. Thus a CT percentage means how far your video will be (at that target filesize) from the highest possible quality.

So like the guides say, 50-60% of maximum quality is acceptable for most people on average (the lower the % the more artifacts you'll get, with zero artifacts at 100%). Moving the resolution slider after having performed a CT is only an approximation, that is why to be sure you'll have to perform a new CT for every different resolution you want to test.

So in short, the CT percentage is a measure for how good the video will look that takes into account the content of the video.

The bpp-value on the other hand is a fixed line that has nothing to do with the content of the video but only looks at the frame dimensions and the bitrate. This value is adequate for a rough estimate, but is useless once you've done a CT, because a CT is much more accurate.

Anyway, that's it in short, I hope this was even a little helpful :).

[manono]

Hi-

Think of it this way. The whole thing at quant 2 is 100%. A one pass at quant 3 is 66.67% (ignoring B-Frames). Quant 4=50%. Many people (including myself), believe that you should aim for an average quant of 3 or less (or a percentage of higher than 66.67%) when working with a nice clean source. If you wanted to use that 640x272, putting on a couple of light smoothers such as TemporalCleaner or TemporalSoften (for temporal smoothers) or Undot (for a spatial smoother), should give you those extra percentage points so you can use the next higher resolution. And they won't make the video any noticeably different than if you hadn't used them. But folio, N_F, and Hiro2k were all giving good advice.

Edit: and Acaila, also. He finished his while I was still typing. :)

[stax76]

On my recent tests I saw blocks on fast moving scenes until I went up to 80%, since then I always use and recommend 80% to be sure to get outstanding quality

[len0x]

Originally posted by Dolemite
On my recent tests I saw blocks on fast moving scenes until I went up to 80%, since then I always use and recommend 80% to be sure to get outstanding quality

even if that gives you optimal resolution 384x288 ? :)

Seriously - this all highly dependent on codec/settings/filters used.
I personally haven't seen blockiness for a while with divx5, although other artifacts can be present (I do lots of b&w movies with 50-60% comp test ratio).

[cereal killa]

Originally posted by manono
Hi-

Think of it this way. The whole thing at quant 2 is 100%. A one pass at quant 3 is 66.67% (ignoring B-Frames). Quant 4=50%. Many people (including myself), believe that you should aim for an average quant of 3 or less (or a percentage of higher than 66.67%) when working with a nice clean source.


Can you explain this a bit more plz?
How does the average quant get set?

[stax76]

even if that gives you optimal resolution 384x288?


should not happen too often using 2 CD's or 3-4 movies each DVD


Seriously - this all highly dependent on codec/settings/filters used.
I personally haven't seen blockiness for a while with divx5, although other artifacts can be present


I spend a lot time coding and only a very few time actually encoding movies. I like to keep things simple and therefore was never much interested in advanced topics. I've learned some advanced things now because I needed to figure them out for the goal of my project but still I won't consider myself as a expert when it comes at filter and codec tweaks. I use DivX default settings and XviD default settings with VHQ 4, chroma motion enabled and Max B-Frames 2. On my tests the blocks were visible pretty similar comparing DivX with XviD using the same settings but XviD looked overall better, maybe it was just imagination because recently I became I'm a GNU geek, wonder why? :rolleyes:

[SiXXGuNNZ]

thanks to all who took part in this thread, you have all been helpful in one way or another.

now I at least understand the comp check and it's purpose :D

[manono]

Hi-

Can you explain this a bit more plz?

Sure. Say you do a one pass quality based encoding at 100%, or a one pass at constant quant 2. That's the best you can get (and again, I'll ignore B-Frames, as they throw a bit of a monkey wrench into the equation). That's your base line best encode. But most people don't do that because they want them to fill one or two CDs. By doing constant quant encodings, you lose file size predictability. And say that a 1-pass at quant 2 is 1000 MB. And say you want the final video size for a 1 CD rip to be 600 MB (with another 100 MB for audio and overhead). Then you have a value of 60% (which will correspond pretty closely to a Compression Test result). To get the average quant for the final size (or for the result of the compress test), divide 60 into 200 and you get an average quant of 3.33. That's where the generally accepted figure of 60-75% comes from. For clean source material, you don't want the average quant to be much over 3 (I don't anyway).

Dolemite likes it around 80%, and I can respect that. I'm convinced that DivX5 uses motion correction, and assigns higher quants to fast motion scenes, which may result in blocks. But I have nothing on which I base that assumption except my eyes. But there's no way to limit quants when using DivX5, the way you can if using XviD or DivX3.11. I limit my quants severely, just so I don't get blocks in fast moving scenes.

len0x said he does B&W movies and gets good results with 50-60%. And I agree. I just finished encoding a movie made in 1934 (Triumph Of The Will by Leni Riefenstahl) and the final result was 55%, or an average quant of 3.7. And it looks great. With noisy older movies you can let the percentages go down.

How does the average quant get set?

For XviD and DivX3.11, the first pass is done with quant 2, so the final video file size you desire is based on that. When using XviD, you can even save the first pass .avi and get a look at that quant 2 encoding, and learn its file size. Since I haven't used DivX5 in a while, I'll just make an educated guess. You fill in the bitrate and the log file has the information to determine how much it gets compressed from a quant 2 encoding. If it still produces an analyze.log file at the beginning of the second pass, the final average quant is near the top of that. But again, that doesn't take into account B-Frames if you use them. And it doesn't take into account if you make changes between the 2 passes (add filters, change the resolution), something that is definitely not recommended. And if I've said anything stupid, I'm sure someone will correct me.

[cereal killa]

thank you manono

I'm still far from understanding how divx(or other codecs) work but at least I'm getting closer :)

[jonny]

@SiXXGuNNZ:

One really important thing is what you pass to the codec.

I make you an example:

I was playing with the Animatrix dvd, targeting for 1 cd.

I've managed to obtain 83% on comp.test (with XviD, bilinearresize@640x272 & some attitional smoothing filters)
Result: 99% artifacts free and 200% details free :D

Doing the same with lanczos@640x272 with no smoothing filters i've got 63% comp.test and a huge overall quality improvements.

I personally try to don't go less than 640x... (expecially with 2.35 ar movies) but the resize filter is also really important.

So i suggest first to focus on the avisynth creation, play the avs file with mediaplayer to decide what filters/resolutions you like.
This is important because in this way you "look at the maximum teorical quality" you can obtain.

After deciding your personal range of resolution/filters on a movie, comp.test is the best way to tweak those parameters.
And finally, as Manono says, consider that remaining in the 60-70% range gives you an average quant of ~3, imho the lower limit for a good encode.

---EDIT---
A quick note about quality based encode:
I think going more that 75-80% of comp.test doesn't give the quality boost to justify the filesize difference from a quant=2 encode (so imo for a really good rip 80% is the higher limit... q=2 encode is an overkill)

[manono]

Hi jonny-

Nice to see you again. You must be spending the summer at the beach or something. You've been scarce recently. Some of what I said before I learned from you, so I hope I didn't say it too badly. But I thought you made at least one point worth repeating. It's easy to get high compress test percentages by putting on the smoothers too strongly, or taking the resolution down too low, and the movie can still look lousy when finished. Just yesterday I gave up a 71% test result with Bicubic to go with a 66% result using Lanczos. And I thought that the extra little bit of sharpness it gave was worth it. And as you said, you can play the .avs just as you can an .avi (or you can do it in the preview window of GKnot), and see how your filters will effect the outcome before encoding.

[SiXXGuNNZ]

great posts manono and jonny

currently I am doing diff comp checks with each resize filter in the newest gknot beta

I am searching now for a list of all 6 resize filters and what they each offer, benefits, etc.

[Kurtnoise]

Hi !!

Just a simple question : Does Comp Check really efficient with B-frames in XviD codec ??

because I haven't the same results with Jonny_Enc tool (thank you by the way....it's really great) and GK....:confused:

[len0x]

Originally posted by Kurtnoise13

Just a simple question : Does Comp Check really efficient with B-frames in XviD codec ??


It'll be the same as in Enc in 0.28.5 which will be out this week...

[jonny]

Hi Manono ^^

Hehe... sadly i'm still @work (and as every year, there are so many things to finish before closing the office for the summer).

But... yep, i must add that beer's party and concerts doesn't help too much the Enc's development :D

I'll return at 100% after the summer ;)

@Kurtnoise13:
Enc's extended method usually add accurancy with XviD+bframes (i think the same mathod will be applyed to GKnot... i was talking with len0x about this in those days)

EDIT:
@len0x: ops... faster than me :)

[len0x]

Btw, does extended method take into account "max bframes" setting of XviD ?
(meaning if it's more than 1 then more frames have to be discarded)

[Kurtnoise]

Thanks for reply guys :cool:

[jonny]

>Btw, does extended method take into account "max bframes" setting of XviD ?
>(meaning if it's more than 1 then more frames have to be discarded)

On my tests extended method was better with maxbf >= 1
It's difficult to make something that take into account of the maxbf parameter, since the threshold make things really hard.
In the first Enc version (not released) i was using a variable snipsize (taking into account the # of bf, but the threshold destroyed all ^^').

Anyway atm (extended method or not) there are some cases where the error is still too high (7%-8% even with 10% of the movie)

The definitive method will be encoding each snip as a separate clip (so no more oversized frames at the end of the snip, discarding only the first frame will be needed in this case).
Raising the snipsize will help too, but i need to finish/test all to calibrate the method (and before speak too much ^^)
I'm still working on this (atm very slowly :D), let me know if you want to make some betatesting or simply take a look when i'll be ready (anyway probably after the summer)