2 Quick Questions About Xvid.

[Dali Lama]

#1. I used "motion search precision" of 5 in the first pass stats file. However, I want to use 6 now in the final pass. Will this work? I mean, will I get that "little" bit more quality or is it too late, because I already used 5 in the first pass?

#2. If I generated a first pass stats file with a previous Xvid binary, but I want to use a newer build, can I use it with the "old" stats file?

Side Note: Is the stats file generated by Xvid the same as Divx 3?

Thanks,

Dali

[Teegedeck]

Originally posted by Dali Lama
#1. I used "motion search precision" of 5 in the first pass stats file. However, I want to use 6 now in the final pass. Will this work? I mean, will I get that "little" bit more quality or is it too late, because I already used 5 in the first pass?
Theoretically, you shouldn't do this - anything you change between passes will make the scaling of the 2nd-pass-bitrate less exact . But as it works to use lumi-masking in second pass only and the difference between MTP '5' and '6' is very small, you could get away with this, too.

#2. If I generated a first pass stats file with a previous Xvid binary, but I want to use a newer build, can I use it with the "old" stats file?
Yes, as long as neither of the two builds used suffered from severe bugs, and as long as the inter-/intra-decsision hasn't been altered between the two builds. If your .stats-file isn't older than two months, it's OK, I guess.

Side Note: Is the stats file generated by Xvid the same as Divx 3?
The format is identical but the content is totally different.

[kastro68]

Just do the first pass again!

It doesn't take long anyway.

Someone told me that the stats file for nandub and xvid are identical...guess he was wrong then.

But you can still use the xvid stats to do a second pass with nandub...but does this take advantage of the quant of 1?

I am too stingy on cd nowadays, and normally do a 2hour movie on 1cd now...i have realised that divx4.12 is better for 1cd rips but you might have to reincode the same movie several times to get the size right.

Xvid is still good for 2cd rips though...but i think it is a bit blockier in high motion scenes as opposed to 4.12


What are others opinion?

[-h]

Someone told me that the stats file for nandub and xvid are identical...guess he was wrong then.

The format is identical, some values are missing in XviD though (luma noise). The important ones (frame size, kblocks, mblocks, etc.) are there however.

But you can still use the xvid stats to do a second pass with nandub...but does this take advantage of the quant of 1?

The Microsoft MPEG4 variants do not support a quantizer of 1, only 2 to 31.

I am too stingy on cd nowadays, and normally do a 2hour movie on 1cd now...i have realised that divx4.12 is better for 1cd rips but you might have to reincode the same movie several times to get the size right.

Define better ;)

Xvid is still good for 2cd rips though...but i think it is a bit blockier in high motion scenes as opposed to 4.12

I've found it to be better than DivX 4.12 in every case I've thrown at it. The frame sizes are smaller in XviD when identical quantizers are used - proof that XviD's motion estimation, AC/DC prediction and other factors are working better than DivX's at this point in time.

-h

[Koepi]

I get better results with XviD as well compared to DivX4 or SBC DivX3.
Even 2hours on 1 CD look still very, very nice - something where you had to tweak around with nandub for hours...(I just remember my 1CD gladiator rip, nice quality - but XviD would do a better job on that, have to get that DVD again just to compare...)

Best regards,
Koepi

[rui]

Currently, i am trying to do a rip of the movie "The Replacements".
I believe the movie doesn't quite reach the 2 hours time, but when doing a divx4 comp check ( i am using this with XviD, the results aren't bad), the only way to get a resolution that would look good is when doing a 2 cd rip.
I think the reason the movie compresses really bad is maybe because of all the football action scenes.
Could XviD do this movie in 1 cd, and still look good?
Remember i am using the divx4 comp check, for the 3 movies i already made, the results it gave me regarding resolution to use when compressing using XviD were good.
Well, i have to try, right? ;)

[Teegedeck]

Koepi,

I hope you're not gonna tell that everyone on your Nandub-forum.
Otherwise you'd be drafted as a moderator over here... ;)

rui,

football scenes seem to be tough; they're also used for some standardized (is there such a word?) codec tests if I remember correctly what I read in c't. Grass, fast motion, colourful shirts...

You should at least use a temporal smoother (2,1), I think. Good luck and tell us what your results look like!

[-h]

football scenes seem to be tough; they're also used for some standardized (is there such a word?) codec tests if I remember correctly what I read in c't. Grass, fast motion, colourful shirts...

High action scenes are a failing of XviD, not to mention the other MPEG4 codecs (DivX 3/4). The decision of when to choose an intra or inter block often goes horribly wrong, leading to the odd situation where I-frames are actually smaller than the P-frames following it (this is the case in high-action anime scenes as well).

This is being worked on.

-h

[kastro68]

I am pretty confident that Xvid produces a much much sharper image than DivX4.12, however I have found Xvid to be much blockier for very high motion scenes...eg the racing scene in star wars or when Froda is running away from the half orcs in the forrest (about 2.67hrs into the movie)


Maybe it is only blockier because the Xvid DSF does not have post processing????? I have yet to examine this though.


Common settings:
kf=250
rate control avg=200000
rate contrl reation=50
rc u/d=5

divX4.12:
min q=2
max Q=6

Xvid:
min q=1
max Q=31

I didn't change the quantizer setting for Xvid because i found it really sensitive to quantizer changes.

I have yet to do much more extensive comparisons, when i have time.

Should i leave Xvid on it's default settings?

AND does anyone know when i should turn on "Smooth quantizer fluctuation"?

[-h]

Maybe it is only blockier because the Xvid DSF does not have post processing????? I have yet to examine this though.

That would definitely play a major part. Try playing the XviD clip with the DivX filter and see what happens.

rate control avg=200000
rate contrl reation=50
rc u/d=5

Just a quick note - those only affect the "1-pass - CBR" encoding mode, they won't change anything else.

AND does anyone know when i should turn on "Smooth quantizer fluctuation"?

It's only really useful if you've set the "I-frame min/max quant" values to something other than 1 and 31. Otherwise, the quantizer doesn't really fluctuate enough to matter.

-h

[rui]

Originally posted by Teegedeck


rui,

football scenes seem to be tough; they're also used for some standardized (is there such a word?) codec tests if I remember correctly what I read in c't. Grass, fast motion, colourful shirts...

You should at least use a temporal smoother (2,1), I think. Good luck and tell us what your results look like!

You can count on it.

I am at work, but when i get home, i will start first pass. With some luck i will do the 2 pass today too. But if not, tomorrow i should have the all movie ready, and then i will fill you guys in about the results.
Right now, i have everything set up to a 2 cd rip, using 608x... resolution (can't remember what was the other value).
But after the 2 cd rip, i will try a 1 cd rip, just to see what it looks like. But i don't expect too good results on the 1 cd rip, because i already tried with divx4, and it wasn't a pretty thing to watch ;)

[rui]

Originally posted by rui
Currently, i am trying to do a rip of the movie "The Replacements".
I believe the movie doesn't quite reach the 2 hours time, but when doing a divx4 comp check ( i am using this with XviD, the results aren't bad), the only way to get a resolution that would look good is when doing a 2 cd rip.
I think the reason the movie compresses really bad is maybe because of all the football action scenes.
Could XviD do this movie in 1 cd, and still look good?
Remember i am using the divx4 comp check, for the 3 movies i already made, the results it gave me regarding resolution to use when compressing using XviD were good.
Well, i have to try, right? ;)

Well, i just did my 2 cd rip of the movie "The Replacements"
This movie is about american football, so it has some hard action football scenes, very hard to encode. I tried once with divx4, and it was a mess.
This one,with XviD, using Nic's latest build, is WONDERFUL!!
I will try tomorrow to make another encoding, but this time using only 1 cd (700MB). This will be a really hard test for XviD.
See you.

P.S. The file size was almost spot on. I targeted for an avi size of 1298046 and got about 1302086. Nothing to be worried about.

[rui]

Well, i just did my 1 cd rip of the movie “The Replacements”, and found out some things.
The test was made using koepi’s latest build, first pass using H.263 and motion search precision 5, second pass using the credits setup, using Modulated quantizer, luminance masking, I-frame locking 1-5, with the Smooth quantizer fluctuation enabled, and motion search precision 5.
First, I still consider XviD superior to divx4.
But…
In the beginning of the movie there is an underwater scene, and after comparing both the XviD rip and divx4 rip, I think divx4 is better in underwater. Maybe because of the “blurrier” picture that divx4 produces, or maybe because I enabled lumi masking, and since the underwater scene is a bit dark maybe the lumi effect stool some bits here.
I just felt divx4 was better. XivD had a lot of small blocks.
Then, in a scene in the jail, which is a bit to the dark side too, XviD again produced much too blocks in here. Maybe because of the lumi masking again??
But, in the rest of the movie scenes, I believe that XviD is a bit superior, not much, but…
Or this movie is just too hard to be made in 1 cd rip :)

By the way, I used divx4 post processing, and didn’t noticed any green or pink colours.

[Koepi]

Hi,
I encoded TWINE with luma masking enabled and 2-3 scenes came out very ugly, those were 3-4 seconds each. I guess it's better to turn luma masking off when using intraframe quantizer lock and quantizer aliasing.

I had no problems with any green or pink blocks either, the overall movie looks great, many details, sharp images,... and this within 570MB, unbelievable.

Rui, please try with luma masking disabled (as I do now), maybe those high blocky are gone then (I can post results in about 1 hour, but since it's another movie I'm trying now it might be that it would have done perfectly anyways ;) )

Regards,
Koepi

[Koepi]

One more answer:

with my actual build (17022002-1) the desired size was exactly 590 MB and the movie came out 589.85MB - 150kb undersized.... so nothing to talk about...

Any ideas what went wrong with your encoding? Any special settings? First pass file from an old first pass with other filters / avisynth script settings?

Regards,
Koepi

[rui]

Koepi, the final file size was almost spot on. No problems there.
The first pass had been done with your yesterday build, so i think that it wouldn't be a problem to use it.
I will try doing the 2 pass with lumi off.

[philippas]

When you use luma masking and the threshold is very high or very low why don't you implement to smooth the image so that you can't spot any squares in the frames ?

[Aktan]

Hi, dunno if this is too late, but I get a lot of blocks with lumi OFF with build feb 16. SO I don't think its becuase of lumi that you guys get a lot of blocks.

[rui]

I did the “The Replacements” 1 cd rip again, this time with no lumi masking.
It improved a little, especially in the jail scene, where it is most dark.
But in the underwater scene, the improvement wasn’t too high.
Still, I must say that using the divx4 comp check, this movie could only reach 29%, when using 1 cd 700 MB and 512x272 resolution. So the problem could be that this movie simply can’t be ripped to 1 cd.
Like I said above, my 2 cd rip, using 160 ABR mp3 audio, it’s very very good.
I think that this movie could be a good candidate to a future Doom9's codec comparison ;)

[Aktan]

Sorry I forgot to post settings...for divx 4 this ANime episode was 24 min avg using a bitrate of 1111. For Xvid, like i said b4, i used build from yesturday using MPEG at 6 with lumi OFF set at the size that divx 4 was suppose to hit (you know divx 4 is undersized =) ). With these settings, I get a lot of notciable blocks during a grey-dark scene. On an interesting note, I get less blocks for these scenes using the same settings with build feb 13.....

[Aktan]

Hi, I just got today's build and saw there is a Modulated Option. Can someone explain to me what this does and when to use it? Also I noticed that you have I-frame quant from like 2-3 while I had it from 1-31. Can someone plz explain what is the difference?

[rui]

Hi
All your doubts can be resolved by going here:
http://roeder.goe.net/~koepi/

Download the pdf file.

Also, i made a reply to the Q&A by doom9, where i copied some information made available by -h regarding the I-frame locking subject

[movmasty]

but isnt the quant 1 to be used only for capture??

[-h]

but isnt the quant 1 to be used only for capture??

It doesn't have to be.. in fact quant=1 is probably overkill for anything - you're pretty safe with quant=2.

-h