Question about Predictive OPV

[archaeo]

I have been reading several previous threads in this forum regarding this topic, but I wanted to see if I understand the process correctly. (Please excuse - noob to OPV - :))
Lets say I run an OPV project (CCE SP 2.66) and it results in "Q value selected=38". As I undersatnd it, A Q value of 38 would probably prompt me to do a multiple pass (maybe 1+1) to get the best results and come in at proper file size. My question is when I go ahead and run this next multiple pass, will I be using that Q value of '38' in my multiple pass setup? More specifically, will I input that particular value (or something around that value) into RB's 'Quality Precision = setting before running the multiple pass? I guess I am just trying to understand how to apply the 'predictive' info generated by OPV to the next run. I know bitrate and other factors will come into play, but I was wondering if the 'Q value selected=' relates directly into the 'Quality precision' setting in RB?

thanks

[Trahald]

The quality_prec basically determines how much of the bitrate available is applied to the flat areas or the complicated areas of the picture. this option uses the bitrate that is given to it.

the 'q' that is predicted has a direct impact on size and is separate from quality_prec. it is a number that is on a scale (determined by the cce people) that says for x amount of quality, this much bitrate is needed (varies by how complicated the video is)

the opv 1-pass is meant to be self sufficient.. you just set the option and let dvdrb do the rest.

[jdobbs]

Here's an explanation -- it applies only to the way CCE does OPV. QuEnc is different (but similar).

In multiple pass Variable Bit Rate (VBR) you define an average bitrate that the encoder must meet. That bitrate is fixed based upon the maximum size of the output, number of frames, and length of the video. The encoder scans the source on the first pass (usually doing a mock encode) to see where the available bandwidth is needed. In the ensuing passes it allocates that appropriate bitrate to each section of the movie (taking from the rich and giving to the poor) so the Q (Quality) is consistent across the movie (and the bitrate meets the average selected). The resulting Q is as high as can be accomplished within the constraints of the output size.

One Pass Variable Bit Rate (OPV) encoding takes a different view. It lets you select the Q you want up front (note: lower is better). The encoder makes a single pass, and as it goes through the source it uses whatever bitrate it required to reach the Q you've selected (within the limits that are set).

The problem with OPV is that since the bitrate isn't controlled -- the output size isn't fixed, and you don't know whether it will even fit on a DVD-5 when its done.

In order to try and correct that, DVD-RB selects a representative sample of the source (usually 1%) during the PREPARE phase, encodes it, and trys to predict what Q is the best that would fit onto a DVD-5 disc (-R, +R, etc). It then uses that Q to do the OPV encode.

The advantage of multipass encode is that it is very accurate in predicting output size. It can use almost every bit available on the output disc -- which could improve quality (slightly).

The advantage of OPV is speed. It can do a complete encode a lot faster. Each pass of a multipass encode takes the same amount of time as a complete OPV encode. So OPV is twice as fast as a 2 pass encode, 3 times as fast as a 3 pass encode and on-and-on. The disadvantage is that Q prediction is a really roughly scaled and it could end up smaller or larger than what you thought (possibly even too big to fit on the output disc). DVD-RB minimizes that possibility but doesn't remove it. The bigger concern for most people is that OPV IS ONLY AVAILABLE on the expensive versions of CCE (you know, the $2000 one you've heard about), so unless you are a video professional you probably wouldn't have a legal copy.

My experience is that you can't really tell the difference between CCE OPV and CCE multipass VBR in most cases.

In QuEnc you don't select a Q in OPV but a bitrate. My experience with that encoder shows that multipass encoded is very much superior to OPV -- so I don't recommend QuEnc OPV unless you're just in a hurry.

[archaeo]

jdobbs, trahald, thank you for your replies. Just to go over a couple of your points:

jdobbs wrote:
One Pass Variable Bit Rate (OPV) encoding takes a different view. It lets you select the Q you want up front

But in Rebuilder's OPV, isn't this automatically selected for you and used for the one pass if you go ahead and encode? Can you tweak Q yourself to something other than the output derived from the 1% sample? I know you can go into Rebuilder's config settings [CCE options] to manually change the representative sampling size from .5%> to better predict overall source, but I thought Q was predicted and set by RB.

trahald wrote:
the 'q' that is predicted has a direct impact on size and is separate from quality_prec....the opv 1-pass is meant to be self sufficient.. you just set the option and let dvdrb do the rest

OK, so 'Q' is separate from 'quality prec' in the settings, it is simply a level of 'quality'. The lower the better. From some posts I've read here, the sweet spot for an OPV Q is somewhere between 18-36. Anything over, you should do a multipass.
And from what I understand, the 'predictive' part of OPV is simply that the Q value indicates what overall quality level can be expected from the encode during a one pass - The 'Q' value isn't to be used as a setting on a multipass, if that's what you decide to do after the OPV?

So if you do an OPV (prepare only) and it indicates multi pass may be a better way to go (like a high 'Q'), do you ever use ANY of the info from the OPV results to get an idea of what you will choose as the settings in RB(ie: VBR bias=, Qual Prec=)?

[Rockas]

@jdobbs
An excellent explanation, as usuall, but don't you think that the prediction percentage would better if set to, lets say, 5%? Don't you think that 1% is too small and it may be the cause to the undersize of previous versions and the oversize of 0.70?
I'm saying this 'cause I've observed other programs like Gordion Knot (yeah I know it's a different type of conversion, but still a conversion, isn't it?) and the default samples for calculations of the compress value are 5%... I know that maybe 10% would be best but that would take too much time. I don't know if putting the value to 5% wouldn't make it slower enought to make it unworth it but it can be tested, have you ever made a try with different values?
How about 2%? wouldn't it give a better prediction than 1%?

Just wondering about :)

keep it UP

[jdobbs]

1% is typical and from the tests I've run it is just as accurate at 5%. Also remember that when you choose 1% you may do as many as 4 passes on that 1% trying to find the right Q... so if you choose, say, 5%, you could be encoding up to 20% of the total (more typically 2-3, though) -- which takes a lot of time, and takes the reduces the single advantage associated with OPV (time).

If someone is dead set on changing it, the sample percentage is adjustable in rebuilder.ini.

But in Rebuilder's OPV, isn't this automatically selected for you and used for the one pass if you go ahead and encode? Yes, DVD-RB selects it for you. If you run 3 click mode you can change it with a global replace in rebuilder.ecl -- but you run the risk of going oversize. When the Q gets into small numbers (15-30) sometimes a single decrement can bring you from being 400MB too small to being 400MB too large. You can only burn one of those two ;). DVD-RB tries to err only on the side of caution.

[Trahald]

yeah.. there were many discussions about this in the early stages of d2sroba ( OPV for dvd2svcd) and it seemed what you gained by going slightly higher % wise never was worth what it cost you in time and usually was no more accurate ( going to .5% was even usable and generally yeilded the same values as 1%)...

[jdobbs]

Yeah. Actually my first few versions defaulted to .5% -- but tylo convinced me to make it 1%, he's the real expert on prediction and helped me write the DVD-RB algorithm long ago.

[archaeo]

thanks all for clarifying ;)

[Rockas]

I'm convinced :)