simple question for CCE people

[EuropeanMan]

In CCE/SP2 is it possible to do a 10-pass encoding? YES or NO.

[buzzqw]

cce sp and sp2 till 99 passes

but usually 3 is enough

BHH

[EuropeanMan]

Do you mean Ninety-Nine? or 9?

[sillKotscha]

why funny?... up to 99 passes is possible...

but as already stated everything over 3 passes is a waste of time

[Boulder]

Actually you can do an infinite number of passes by re-using the old vaf file.

[spanky123]

Actually you can do an infinite number of passes by re-using the old vaf file.

Lol

When you say 'anything over 3 passes is a waste of time', does that mean we should be selecting pass = 3 or pass = 2, coz of that extra analysis pass that CCE does at the start ?

[Boulder]

If you use the adaptive quant matrix feature, I recommend doing the vaf file creation pass + two passes. Otherwise you might get under/oversizing and also the feature needs at least the one extra pass to do its work properly. If you don't use it, you probably only need vaf+1 pass.

[kolak]

Adaptive quant needs more than 2 passes ( it doesn't work at all when you do vaf+1 pass).
I wouldn't recomend for good quality vaf+1 pass. It's not enough.
vaf+3 passes is OK.
The power of CCE are many passes (and speed).

[Fishman0919]

Adaptive quant needs more than 2 passes ( it doesn't work at all when you do vaf+1 pass).
I wouldn't recomend for good quality vaf+1 pass. It's not enough.
vaf+3 passes is OK.
The power of CCE are many passes (and speed).

Adaptive quantize matrix works with vaf+1 pass. 1 extra pass for high bitrates is needed sometimes for sizing

vaf (1) + 1 pass is more then fine for 99%, some real low or real high bitrates may need 1 extra pass.

The idea behind more then two passes, in short, is to take a little from here where it may not be needed and put it over here where it MIGHT look better.

It's not a magic wheel where the more you turn it the better it gets, the enocoder can't magicaly pull bits out of nothing. CCE gets ALL the info it need to encode the video in the first pass (.vaf pass) and applies it in the second pass just like every other encoder. It simple math (well maybe not real simple) ... the 1st pass gives the encoder the info it needs and then it distribute the bits to where they are needed on the 2nd pass. The 3rd, 4th, 5th and so on are really just 2nd and 3rd guesses and not every guess is going to be better... In some test 6 or 7 passes look worst then 2 or 3.

[kolak]

I said: adaptive quant doesn't work at all with 1 pass.
Sorry this not fully true.

Vaf+ 1 pass may change mattrixes, but additional passes will improve matrixes more (vaf+2 passes is min. to allow for this option really works- this is form Cinema Craft engineers and confirmed by my tests).

The same with number of passes. CCE will improve overall quality within each pass (the more passes the lower incrementation of improvemnt).
I don't think that 2, 3... pases are just a guesses. CCE tries to lover the average quantization level.
How many passes are you gona use depends on source and the use of the file. If it's only encoding for home use (often the source is mpeg2 already) than it's fine to use 1 pass.
If you have uncompressed source than 3 passes will always look better than 1 (especially eg. on fades).

I don't really care about number of passes, because my machine does 1pass (from uncompressed source) 10 times faster than RT, so I use 3 passes as a standard:)

To avoid aswer for your next question, file server is build on 20disks (over Areca raid controller) and the network uses 10Gbit Infiniband connections.

[Fishman0919]

I said: adaptive quant doesn't work at all with 1 pass.
Sorry this not fully true.

Vaf+ 1 pass may change mattrixes, but additional passes will improve matrixes more (vaf+2 passes is min. to allow for this option really works- this is form Cinema Craft engineers and confirmed by my tests).

The same with number of passes. CCE will improve overall quality within each pass (the more passes the lower incrementation of improvemnt).
I don't think that 2, 3... pases are just a guesses. CCE tries to lover the average quantization level.
How many passes are you gona use depends on source and the use of the file. If it's only encoding for home use (often the source is mpeg2 already) than it's fine to use 1 pass.
If you have uncompressed source than 3 passes will always look better than 1 (especially eg. on fades).

I don't really care about number of passes, because my machine does 1pass (from uncompressed source) 10 times faster than RT, so I use 3 passes as a standard:)

To avoid aswer for your next question, file server is build on 20disks (over Areca raid controller) and the network uses 10Gbit Infiniband connections.


You can do as many passes as you like... but like I said "It simple math".... all the info is got in the 1st pass.

I helped test www.autoqmatenc.com (http://www.autoqmatenc.com) and built my own encoder. More passes are really just guesses (slight adjustments for something not done right the first time) and in the case of CCE, it's not always going to be better.

Edit: spelling

[Sir Didymus]

...
CCE will improve overall quality within each pass (the more passes the lower incrementation of improvemnt).
I don't think that 2, 3... pases are just a guesses. CCE tries to lover the average quantization level.
...

I strongly object to the first part of your statement. From a mathematical point of view the process is not linear, nor continuous.

Whenever the encoder have already reached (or it is near to reach) the minimum average quantization level - given the available bits for this job, any further attempt to gain more quality is exactely a guess (as correctly argumented by Fishman0919) that is most time resolving into a loss of time...

Me too, I did many tests on the matter, based on SSIM metrics, convincing (myself...) that CCE gets is optimal performance with two ot three passes maximum (including VAF generation).

Any further passes is exactely a guess: it may improve a little bit the SSIM value, but it may equally lead to worse results...

Cheers,
SD

[spanky123]

Sorry. What's all this stuff on
'cce adaptive quantizer matrix' ?

I can't find this anywhere in CCE SP2. Iirc, I remember seeing it in an older version of CCE.

[kolak]

I'm still saying that you need at least vaf+2 passes to get the best form CCE-SP:)

[Fishman0919]

Sorry. What's all this stuff on
'cce adaptive quantizer matrix' ?

I can't find this anywhere in CCE SP2. Iirc, I remember seeing it in an older version of CCE.

It's the same from CCE SP as in CCE SP2.

When CCE SP/SP2 can't keep the minimum bitrate. It is caused by the lowest quantization scale not able to raise the bitrate under a single quantization matrix. When this happens, CCE SP/SP2 automatically creates a new matrices cutting the value of each element in the matrices by half, quarter and so on, and applies the new matrix to produce a picture with less distortion and a much higher bitrate. This switching of quantization matrix can occur at every GOP.

In short... if you are using a matrix is too low for the bitrates being encoded, CCE SP/SP2 automatically adjust the matrix to be better suited for the high bitrate.

[spanky123]

Ok. Thanks.

I see you can disable this. You shouldn't though, right ?