Xvid Encodes Fasssster than 4.12 :)

[kastro68]

It does, well at least until B frames are introduced...but i am content atm

[MaTTeR]

I know! I just tried the latest build last night and it's absolutely wicked fast on my dual PIII system.

[Franko30]

Originally posted by MaTTeR
I know! I just tried the latest build last night and it's absolutely wicked fast on my dual PIII system.


Which one would that be? Koepi, Nic, Umaniac? Of which date? I'm using the Kopei Feb 17th build and ddn't notice any speed improvement over previous builds...

Hope you enjoy your Jack Daniels from the beer thread :D

Cheers

Frank

[Ripe73]

Which one would that be?

It must be Nic's, i think

[Franko30]

But Nic's binaries ain't got the Quantizer modulation yet - too bad - going back to Koepi's binaries of Feb17th - 8 great encodings already done with it!

Frank

[rui]

Originally posted by Franko30
But Nic's binaries ain't got the Quantizer modulation yet - too bad - going back to Koepi's binaries of Feb17th Frank

Same here.;)

But, this isn't the first post i see saying that the latest Nic's build (compilation of the latest -h CVS) is a little faster.
Maybe Koepi could compile the latest -h CVS code and introduce the modulated quantizer in it.

[rui]

By the way, Franko30, what were your settings in your rips using Koepi's build?
I never messed with the "normal frames" ( i believe they are called inter frames) quantizer, so i leave them at 1-31 default.
I only modify the intra frames quantizer locking, using 2-6 for 1 cd and enabling the smooth quantizer fluctuation.
Could i get better results by messing with the inter frames quantizer?
If yes, could you give me some values to start with?

Thanks.

[sierrafoxtrot]

@rui

hey. from what i've gathered, you can set min 1-2, max quant to 5-6 for 1CD, if you use lumi masking, to avoid over-aggresive compression in the darker scenes. other than that, i haven't found any great difference, and setting the interframe quantizers too low can mess with filesize predictability. i've only ever changed them a couple of times when doing a 2CD rip of saving private ryan. it didn't exactly rock my world too much. sorry for the waffly reply.

good luck testing anyway. you seem to be responsible for a lot of feedback. that's really good. :p

[MaTTeR]

I used Nics build last night.

Ok so maybe I had to much Jack last night and the progress bar looked faster:D Not! It's a little faster but I'm not sure how much as I haven't had time to thoroughly test it.

[rui]

@sierrafoxtrot:

Thanks for the tips :)

[Franko30]

Originally posted by rui
By the way, Franko30, what were your settings in your rips using Koepi's build?

I only modify the intra frames quantizer locking, using 2-6 for 1 cd and enabling the smooth quantizer fluctuation.
Could i get better results by messing with the inter frames quantizer?
If yes, could you give me some values to start with?

Thanks.

Hi rui,

sorry for the late reply - but yesterday evening, for a change, I didn't spend my time in front of the computer, but did some old fashioned reading (Philip Kerr: The Grid). And sometimes even I have to work ;)

For the settings: I noticed that my encodings almost never use a Quant higher than 9 - except for credits. So instead for 1-31 for the Inter frames, I use 1-14 (as this would get used on credits and 16 doesn't look good enough for me, even on credits) and the I-frame Quant lock is set to 1 to 6 (Quantizer fluctuation smoothing enabled).
I chose 1 to 6, because my encodings are done from analogue source material and the average Quant is around 3-4 but seldom above 8-9, so I figured 6 would be a good setting for I-Frames - maybe I'm going to try 5 - but this might affect filesize predictability.

Does anyone have any experience with "too tight" Quant settings?

Frank

Edit: And BTW, my encodings result in only a 2 to 150 kb difference to the given filesize (that's video only).

[rui]

Hi Franko30

Thanks for your reply. I like to read too, currently i am finishing Lord of The Rings, the third book. Too bad that i still have to wait for next Christmas to see the next movie :(

But, anyway, after reading your post, something came to my mind.
This is sort of a newbie thing to ask but here goes:

I always thought that after doing the 1 pass the codec would then decide automatically what quantizers it would give to every scene in the movie. But, if I understood your post correctly, then the codec will only choose what quantizers to apply of those that I indicated (in your case 1-14), correct?
But lets say, that you and I are ripping the same movie. I choose the default inter frame quantizers (1-31), and you choose 1-14.
Now imagine that the codec reaches a certain scene in the movie and decides that for that scene he needs a quantizer value of 8. Both my movie as yours get equal quality, right?
Then, in another scene, the codec, after doing 1 pass, decides that the scene only needs a quantizer value of 20. My movie will get shorter than yours because in your case the codec could only choose 14, correct?
So, where is the codec going to put the extra bits he saved in that scene, in my movie, to reach the 700 MB?

Thanks in advance.

[Franko30]

Originally posted by rui
Hi Franko30

Thanks for your reply. I like to read too, currently i am finishing Lord of The Rings, the third book.
Then, in another scene, the codec, after doing 1 pass, decides that the scene only needs a quantizer value of 20. My movie will get shorter than yours because in your case the codec could only choose 14, correct?
So, where is the codec going to put the extra bits he saved in that scene, in my movie, to reach the 700 MB?

Thanks in advance.

Well - I finished book 3 before the movie part 1 came out and then started the book all over again - I always keep forgetting what happened in book three (especially the end).

The question about "where go the saved bits":
I don't think your movie of the example might get longer, maybe a little smaller. It all depends: I guess the quantizer fluctuation control brings it all back to equal filesizes over a long enough given timeline (and remember: on a long enough given timeline, we're all dead...) :D

I think this would be a good question for Nic or Koepi, as they introduced this thing about quantizer settings.

Good night

Frank

[rui]

Franko30, thanks for your reply.

Maybe Koepi and/or Nic care to step in ? ;)

P.S. This forum must have the biggest number of German guys i have ever seen!:D

[Teegedeck]

...just citing suxen_drol on how XviD's bit bucket works:


if there is an overflow 12.5% of the bits are given to the next frame
if there is an underflow 25% of the bits are taken from the next frame
and we never give or takeaway more than 1/2 of the next frame size.


Of course if could happen that bucket doesn't empty before the end of the movie (for example if it would end with a very darks scene when you use lumi - but I haven't seen such a movie, yet) - I wonder whether the credits-algo allows to spend remains of the bit bucket, too?

[-h]

Of course if could happen that bucket doesn't empty before the end of the movie (for example if it would end with a very darks scene when you use lumi - but I haven't seen such a movie, yet) - I wonder whether the credits-algo allows to spend remains of the bit bucket, too?

The bit-bucket doesn't really work that way anymore..

I had to change it to fix Doom9's undersizing issue. Now, if the overflow is greater than 75% of the desired frame size, 75% of the decimated overflow is added to the desired size. By "decimated overflow", I mean 12% / 25% of the value mentioned above.

It.... should fix it.

-h