CCE passes with DVD-RB Pro?

[Chip718]

After seeing the impressive quality of DVD-RB I had to donate and get the Pro version. The Rockas Installer automatically set my CCE SP passes to 2. I am not really familiar with CCE's settings, but if I upped the passes would I get a even better quality picture?

Thanks for any info.

[adam]

Probably not. A setting of 2 passes in CCE really means 3, because the creation of the .vaf file is the real first pass. In my opinion, 3 pass VBR is probably overkill as it is.

Beyond the second pass you don't gain much at all despite what all the 5, 6, 9 passers tell you.

If you are using low bitrates than consider more passes and as always do a couple test encodes and decide for yourself.

[tigerman8u]

IMO anything over 2 passes is a waste of time. on many tests I've done my eyes have seen no quality difference

[manono]

adam, DVD-RB doesn't use CCE's convention for counting the number of passes. 2-pass for DVD-RB means the .vaf creation plus one additional pass.

IMO anything over 2 passes is a waste of time. on many tests I've done my eyes have seen no quality difference

I can only assume you mean the .vaf creation plus 2 more passes. You may or may not be able to tell with your eyes, but the Q-Scale graph as seen in the VBR Bit Allocation Window continues to smooth itself out for several more passes. That is, the quality from GOP to GOP continues to level out and to fluctuate less. For me, that means better overall quality comes with more passes.

[hrlslcbr]

But the important thing is what you see with your eyes. You won't have a graph telling you that the video has good quality while you're watching it, will you?

[adam]

I didn't know that about CCE's naming scheme. It does make more sense...unless you're just really used to CCE.

I've checked the bitrate and quantization levels after each additional pass and the differences after the 3rd pass are so ridiculously minute that I can't believe anyone could actually notice a visual difference. Even from the 2nd to 3rd pass the differences seem neglible. Also I don't see the Quantization levels normalizing with additional passes and I don't know why that would be indicative of higher quality. If all you want is an even Q than do OPV with a very high bias.

[bmiller]

I've been using 3 pass,which is in effect2 pass cce.Great quality video.

[manono]

Hi-

If all you want is an even Q than do OPV with a very high bias.

That's exactly what I do want. I want that quant scale graph to be as straight a line as possible, complex scenes with the same quality as static scenes. This isn't usually possible, of course, as very complex scenes will often run up against the max bitrate set and cause the quants to rise. However, I also want the final DVD to be 4475-4480 MB in size, and OPV isn't too good for size prediction. And I think you must mean a very low bias, since true VBR encoding comes with a low Bias. I have mine set for 0.

Sure, call me anal.

[adam]

Yeah I did mean low... You can still run the OPV pass and then a second pass to make the adjustment to the exact size. That will give you about as close to a constant Q as you can get with CCE.

But I still don't see the point.

I want that quant scale graph to be as straight a line as possible, complex scenes with the same quality as static scenes.

You get the exact opposite with a constant Q. Quantization does not equal quality, it is basically the amount of data discarded. In normal VBR encoding the quantization fluctuates to maintain more or less equal quality. Constant Quantization would mean that the same amount of data would be discarded from each scene regardless of the resulting quality.
The actual quality will vary more than in normal VBR mode.

[manono]

You get the exact opposite with a constant Q. Quantization does not equal quality, it is basically the amount of data discarded.

You couldn't be more wrong, in my opinion. If it were a fixed amount of data being discarded, then complex/larger frames would have better quality than static/smaller frames, as less would be discarded relative to the original frame size. Rather, approaching constant quant means that the same percentage of data is discarded from each frame, leaving all the frames with more or less equal quality.

In normal VBR encoding the quantization fluctuates to maintain more or less equal quality.

It fluctuates because of the settings that were used either for the commercial DVD, or for the backup encode, or both. The bitrate curve fluctuates (a lot), if done right, but in my opinion, the less the quant curve fluctuates, the better. Yes, I know that this idea goes against a lot of advice floating around out there. Have you made AVIs before? Do you not agree that the best quality comes from making constant quant encodes, and that the only reason for 2-pass encodes is to fit the AVI to a given file size, such as 1 or 2 CDs? Maybe not. With my AVIs, I do fit my encodes to a fixed file size, but I severely restrict the quants when doing so, so that it approaches constant quant encoding. When using CCE for DVD backups, I try and do the same thing with my settings.

[adam]

You couldn't be more wrong, in my opinion.

Never underestimate how much more wrong I could be.

No I don't ever encode to avi so at the very least we are coming from different perspectives.

approaching constant quant means that the same percentage of data is discarded from each frame, leaving all the frames with more or less equal quality.

Yes I agree that a constant Q would mean a constant % of data discarded uniformly. I think where we have a discrepancy is in our definition of "quality." The whole point of variable quantization, as I'm sure you know, is to exploit the human eye's high frequency roll off. You can use a larger step size (more quantization, aka more compression) on high frequencies without any visually noticable difference as compared to low frequencies.

So applying constant quantization compresses everything equally, and from a mathmatical standpoint that means uniform quality, but that just means that the coefficients are rounded off the same.

My definition of quality involves actual perception after the coefficients have been inversely quantized and if high frequencies compress better than low ones I think I will visually see more uniform quality if I compress the high frequencies more instead of equally. More importantly I should achieve an overall lower Q since I exploit the compressibility of high frequencies, and assuming my encoder quantizes conservatively enough I can't imagine I'd ever see any quality variation regardless.

I still just don't see a point to constant quantization. You can compress everything by the same amount but that doesn't mean everything will compress equally. Basically I just think its an overly conservative method of encoding that values a level of consistency, which you can't "see" anyway, over an overall better compression (lower average Q.) I must say that constant Q would basically eliminate all mosquito noise, which is good, but of course you can accomplish the same by just tweaking your settings intelligently.

We're both obviously anal about this, no one even uses the word quantization unless they are anal about tweaking their encodes down to a nuance. I guess we just have different priorities.

[manono]

Hi again-

Basically I just think its an overly conservative method of encoding that values a level of consistency, which you can't "see" anyway

I can't disagree with you on that one. It pretty much boils down to encoding philosophies. I'll grant you that unless you're pausing frames in the middle of an explosion, for example, the chances are good that you can't tell that it doesn't look as good as the rest of the movie when watching at normal speed. I just like to know that I've done my best to make the quality as consistent as possible.

More importantly I should achieve an overall lower Q since I exploit the compressibility of high frequencies

Is that true? I'm asking because I've never tested. It seems somewhat counterintuitive to me. You compress the relatively few complex scenes by a lot more than I do, and you compress the more common static or near static scenes by a very little bit less, and it should even out, it seems to me. I'd think that there would be very little difference in the 2 average quants. But like I said, I just don't know, and until I run my own tests, I'll take your word for it.

Speaking of filtering high frequencies, as perhaps you know, CCE does quite a bit of additional high frequency filtering by default. Have you seen Doom9's recent CCE 2.70 guide?

http://www.doom9.org/mpg/cce270.htm

Scroll down a little bit more than halfway, and find the picture with the black on top and the green on the bottom. Does your CCE look like this? It should, in my opinion. Having the Low Pass Filter (and the Vertical Filter) enabled is bad news if you value sharpness and detail in your encodes. There are some things about which I disagree in his guide, but I think he's right on about that pic.

But one of the worst things about CCE is having the Standard Matrix as the default. You can't change that in CCE 2.50, and the more recent versions have only the MPEG Matrix as a marginally better available matrix. If you care about the quality of your encodes, you can throw all the bits at it you want, 4000, 5000, 6000, but your encode is still crippled, as far as I'm concerned. It does allow the option of adding matrices to the list, and I highly recommend the study of the high bitrate matrices used in commercial Hollywood DVDs, and using some of them whenever possible. That Standard Matrix just filters away way too much high frequency detail. So, unless you're careful, high frequency filtering just gets compounded by CCE, and the resulting encode can't look anywhere as good as the original DVD. It does have a couple of very good low bitrate matrices available for encoding such things as low bitrate extras. Food for thought about a subject I'm interested in. I enjoy talking with you, adam.