DivX vs. XviD compressibility

[JohnMK]

I've noticed something a bit distressing. Doing compressibility tests with XviD at its most-used settings (VHQ4, ME6, chroma optimizer, etc.) results in compressibility about 10% higher on average than DivX 5.1 or the 5.1.1 betas. I always use B-Frames, and always DivX on the Standard setting. Slow/Slowest actually decreases compressibility slightly further!

Any theories? Is XviD really better at compressing movies than DivX? This means I can often use a notch or two higher resolution if I choose to use XviD, but I'd rather use DivX since it's so much faster. Hmmmm what to do, what to do.

[Danzel]

My recomendation: Dont use compressibility tests :P

As you've seen they think that slow/slowest compresses worse than standard, which sounds certainly not right....

Try compress the movie for the size you want it at the resolution you want it and see what it looks like.

And the speed differences: those most used settings on XviD are pretty much the Highest settings (Note: Pretty much). So you could expect it would be slower than the standard in DivX 5.

Disabling / lowering your VHQ setting in XviD will give much faster speed.

Danzel.

[stax76]

My recomendation: Dont use compressibility tests :P

As you've seen they think that slow/slowest compresses worse than standard, which sounds certainly not right....

Try compress the movie for the size you want it at the resolution you want it and see what it looks like


but how do you know you need 500 MB or 1000 MB, how do you know how good the movie compresses before you actually encode? Most people here encode a movie several times just to play around and to learn something, test new codecs etc. but some people just need to get the job done. Comp. tests are useful to give you a raw idea, nothing more, nothing less


Any theories? Is XviD really better at compressing movies than DivX?



for testing and comparing compression and quality I agree with Danzel to encode and look what it looks like, different codecs, codec settings and programs produce different comp. test results. I bet on DivX when it comes to compression but I must admit I rarely test codecs

[curna]

Any theories? Is XviD really better at compressing movies than DivX? This means I can often use a notch or two higher resolution if I choose to use XviD, but I'd rather use DivX since it's so much faster. Hmmmm what to do, what to do.

Well, here is my theory.

A compressibility test it's only a ratio.

MB of video encoded with your settings / MB of vido encoded at codec full capacity.

If this ratio is lower in Divx could be because, with new versions, at full capacity the file is bigger than Xvid at full capacity.

In other words, if you take the same video and encode it using the full capacity of Xvid and divx, the divx version will always be bigger. Maybe because it tries to keep more details and uses more MB for them (with the improvements in the new version).

However, this has nothing to do with quality at your settings. First, becuase quality "is not lineal" (10 MB bigger doesn't mean always a same perception of quality improvement), and second, because you cannot say that at full capacity the 2 codecs performs exactly equal in quality (so that the only difference is the size), but some of them could .

In conclusion:

If you want to use Divx I would keep on using it. However, with this new version and the changes made to it, you should first investigate a little bit which ratio in the compressibilitty test is good for you. With previous versions of Divx it was said that good values for the compressibility test were 60-80. With this new version, to have a similar quality, I've heard that it should be around 50-70. But the best idea is that you do your own tests.

The important thing is that a lower compressibility test with different codecs doesn't mean lower quality.

[Soulhunter]

Doing compressibility tests with XviD at its most-used settings (VHQ4, ME6, chroma optimizer, etc.) results in compressibility about 10% higher on average than DivX 5.1 or the 5.1.1 betas.
Does "most-used settings" include MPEG matrix ???

AFAIK with MPEG matrix the XviD encode should be same size or even bigger than the DivX encode... :confused:

What settings have you used with DivX 5.1.x ???

Have seen that that DivX 5.1.1's Slowest mode + Slow Psy. mode increases filesize @ Q2, so that would also mean it decreasing the compressibility... ;)

Bye

[cordraconis]

Originally posted by Soulhunter

Have seen that that DivX 5.1.1's Slowest mode + Slow Psy. mode increases filesize @ Q2, so that would also mean it decreasing the compressibility... ;)

Bye
I just noticed the same thing a few days ago.
I was doing a Q2-pass to calculate/estimate the compressibility of "The Matrix Reloaded", and I used Slowest + Slow. Psych.
After about one hour I checked the estimated filesize to be over 4 Gig! :eek:
I aborted the process to see if I didn't use Q1 (a mistake I made a few times in the past:rolleyes: ), but I found nothing.
Then I put the speed slider on Standard, and bingo! After 30 min the estimated filesize was around 2 gigs.

I don't know how it works, but I think the quality slider affects the motion search intensity.
If you use it in Nth-pass, the codec will look for the best bitrate distribution for the given amount of complexity (motion). If you use it in Quality-based modus, the codec will detect more (finer)motion and because filesize is not important, it will allocate more bits to it.

... but I never thought it would make twice the difference in filesize.
If my last pass finishes, then I'll try to do a Q1-encode with Slowest. If my guess is correct, then the final size should be more than the VOB-files! :eek:

[Soulhunter]

Yeah... Ive posted about this subject here !!! (http://forum.doom9.org/showthread.php?s=&threadid=63690&perpage=20&pagenumber=3)

Bye

[DigitAl56K]

Actually either Psy mode will increase the filesize if you are doing a quality-based encode purely because they manipulate the texture in a way that typically requires more bits at any particular quantizer.

I don't particularly recommend using PVE in combination with 1P-QB mode unless you have a very specific reason for manipulating the texture.

[Soulhunter]

@DigitAl56K

But does the compression test not also use QB encoding... ???

So when you want to use PVE for normal multipass, how would you make a compression test then... ???

I'm a bit confused now... :confused:


Bye

[cordraconis]

Originally posted by Soulhunter
@DigitAl56K

But does the compression test not also use QB encoding... ???

So when you want to use PVE for normal multipass, how would you make a compression test then... ???

I'm a bit confused now... :confused:


Bye
Well ... "don't" would be my advise.

After your earlier comment that you already discussed the issue, (Sorry, but I don't read all the threads :) ), I tested the scene where Neo fights the 3 agents that come banging through the metal door.
1085 kb/s, DivX 5.1, 5 pass multipass.
I tested with Standard and Slowest speed, and combined each with Qpel and Srtong PV. (both or without.)

Using Qpel adds a little bit more detail and grainyness tot the picture, and the grainyness is most obvious in the blurred background, for example when they zoom in on Neo's face and the 2 people after him turn blurry.
I remember reading something about Qpel not working in Slowest mode, but there are definately grains when doing this.:confused:

The effect of PV is that the macroblocks in the background dissappear almost totally. No "musquito noise" anymore. Using Slowest mode even further reduces this. Also more details in the low-motion scenes.
The arm movements of the agent during the fight, are more blockier than without PV. The blockyness is slightly reduced with Qpel, probably by more accurate positioning.
This concurrs with what DigitAl56K said: If PV increases the amount of bits for a texture, then the extra bits are probably taken from the background motion-information, resulting in more details for the textures, and less movement in the background, resulting in a bonus stability.

So if the PV redistributes bits in an encoding, the effect on QB encoding would be that the parts where the bits were taken away, will still be encoded with constant quality, so the net result would be that some textures recieve excess bits, and hence the total filesize is higher.

Therefore (as I stated before), PV should not be used in compression tests, but definately in the multipasses.

Hope I figured this one out without mistakes! (DigitAl56K ?)
Interesting to know, dough.

[dTb]

The compressibility test is a very valuable tool but imo it shouldn't be used for comparisons between codecs etc.
I use it for every vid I compress but I keep it simple, q2, b-frames, standard p/q and nothing else. This way you can use it to more accurately judge the compressibility of a movie, if you add pve in for a movie here and there it becomes harder to pin down percentages your happy with.
Like someone mentioned, with divx 5.1 with fast pve and slowest p/q I've revised my lower limit down from 60% to 50% but note that I don't use the pve in the compress test because I want it's results to provide a standard. The effect of pve is something to figure in after the compress test is done.
If I was to encode with Xvid, I wouldn't use my 50% cut off because for all I know xvid may compress better and allow an even lower cut off, and this isn't even considering the differences in a q2 encode between the two codecs which will result in different predicted filesizes.
So my advice, keep it simple, use a standard method and don't use it for things it wasn't intended for.