~11 hours on DVD-9: quality improvement?

[moviefan]

Hey guys,
I have about 11 hours, actually 10h47m of video material that I want to encode with CCE 2.7 SP Trial to 7.95GB (DVD-9). I have tried several settings using AviSynth in combination with Low and Vertical filter in CCE and all this in Full D1 and Half D1, but I'm not quite content with the quality. I calculated the average bitrate of the video which is about 1300 kbit/s. I know this isn't a lot, but I'm not expecting full DVD quality. In scenes with a normal amount of details the picture looks good, but sometimes, and this is really weird, I have massive macroblocking... and I'm talking of really huge blocks - other scenes have almost no blocks. Somehow Full D1 sometimes looks better than Half D1 because it's more precise.

So that's what I can tell about my experience with this video and the settings I tried. I hope you have further ideas how to improve the quality.

[JoeShrubbery]

Well as it's video footage, if it's interlaced then try some deinterlacing before encoding. It's not ideal, but when you're dealing with bitrates this low you have to make compromises.

Other than that, I'd just suggest you use a second disc and double your bitrates. 1300 is less than I like to use for my mpeg4 Xvid encodes, I can't imagine it being pretty trying to use with mpeg2.

[smok3]

well, i saw some 'impressive' clips made with vlc at 1500 kbps today, so maybe try that (or mencoder).

[Sagittaire]

Perhaps possible with 2.35 movie with 4/3 DAR and with source really compressible.

If your source is DV Pal full 720*576 resolution (generaly very incompressible source) it's really impossible with good quality.

[jshumate]

Well, you are approaching VCD type bit rates, so you are getting VCD type results.

I have a suggestion though. I saw an example once where more or less a guy wanted to do something somewhat similar to your task. He used variable bit rate with CCE and did something like 9 passes and it turned out pretty good. If you are using constant bit rate, VBR might work out a little better for you, but be careful on the maximum bit rate settings because if you go too high, your video may be bigger than you think. Do at least 3 passes, maybe more, to improve quality. I thought that 9 passes was crazy, but the results really were pretty good.

[YouDontKnowMe]

make a xDVD with TMPGenc Xpress and than burn in to dvd I mean 352x288 is great with 1300kbits ;-)

[Mug Funky]

if you're going half-height, there's no reason to not use mpeg-1... it's DVD compatible, where MPEG-2 at CIF resolution is not.

352x480 or 576 is DVD compliant, just make sure you don't have subtitle tracks as DVD players will probably bugger them up.

[shirohamada]

you can try 480x480 ntsc. svcd res.
not compliant.
you need to patch em with dvdpatcher if your authoring tool won't accept.

in my eyes, not much difference between full res and 480x480, on sdtv. but certainly a big difference with 352x480 and 480x480.

my record is 7+ hours on dvd5. using quEnc witl qlb matrix constant Q2. anime material. can see qlb artifacts, even at Q2.

[Chainmax]

I was actually going to post a similar thread as I want to make a DL disc with 13 episodes of a series (~13hrs). The total bitrate for audio + video will be ~1550kbps and I wanted to ask a few questions that are interconnected:


1) I was wondering if upsampling the audio (ursamtl's guide with free software gives pretty impressive results, at least to my untrained ears) to 5.1 and using 384kps AC3 was worth it of if I should just go with 192kbps stereo AC3 and save as much bitrate for video as possible.

2) Taking into account the choice made in 1), would you try full D1, Half D1 or VCD resolution MPEG2 (all of which are DVD compliant according to VideoHelp (http://www.videohelp.com/dvd))?

3) Taking into account both previous questions, which matrix would you use?

[danpos]

@Chainmax

IMO:

1) Don't do upmix to 5.1 and let more room to video;
2) This depends of original horizontal resolution (use the fitcd for to do the correct resize);
3) if you don't want to bother with this question, use AQE (http://www.autoqmatenc.com) with QMatOp ON and it will choose optimal QMAT for your target bitrate (some results (http://www.vmesquita.com/forum/index.php?topic=7099.msg47603#msg47603) of mine using AQE ;) ).

Regards,

[Chainmax]

I guess upmixing to 5.1 isn't such a good idea, after all the audio can be left at stereo and still enchance it using just the fLfR plugin.

As for the second answer, I do use FitCD with 2-block overscan in mind, but are you sure it depends on just the original resolution and not the target video bitrate?

Also, I'd rather ask opinions than to just use an automatic program make the decistions for me. I know that for 352x240 and 1358kbps, AVAMAT7 would do nicely, but I don't know what matrix to use for Half-D1 or Full-D1, or if it's even worth using an ultra low bitrate matrix since I assume the matrix will wipe out much of the detail that using full resolution would keep.

[Boulder]

I would use half-D1 with the QLB matrix. If you use CCE with adaptive quantization, you might do with a lower bitrate matrix as CCE can halve or quarter the matrix depending on the GOP in question.

I wouldn't go for VCD resolution, it'll make things too blurry IMO.

Keep the audio at 2ch, no need to upmix as your amplifier can probably do that for you well enough - at least mine does a good job with DPL and DPLII.

[Chainmax]

Is this:

http://img482.imageshack.us/img482/9374/qlbv12xg0.png (http://imageshack.us)

the QLB matrix? Also, for a 4:3 movie, Half-D1 with 2-block ovescan would be Lanczos4Resize(304,448).AddBorders(24,16,24,16), right?

[Boulder]

Nope, that's not QLB.
08 16 19 22 26 27 29 34
16 16 22 24 27 29 34 37
19 22 26 27 29 34 37 39
22 22 26 27 29 34 38 42
22 26 27 29 32 36 40 50
26 27 29 32 36 40 50 61
26 27 29 35 40 50 59 75
27 29 35 40 50 59 75 89
16 17 18 19 20 21 22 23
17 18 19 20 21 22 23 25
18 19 20 21 22 23 24 26
19 20 21 22 23 24 26 28
20 21 22 23 25 26 28 29
21 22 23 24 26 28 29 31
22 23 24 26 28 29 31 34
23 24 25 28 29 31 34 38
Check the resizing with PARanoia, I trust that program nowadays the most:)

[Chainmax]

Could you provide me to a direct link to the CQME matrix? It's incredibly annoying to fill 128 boxes all the while double-checking you entered the correct number...

[Boulder]

http://www.saunalahti.fi/sainki/QLB.xcm

[Chainmax]

Thanks :).

[Chainmax]

The matrix you linked me to:

http://img67.imageshack.us/img67/9432/qlbsw3.png (http://imageshack.us)

has slightly different numbers than the ones you quoted. By the way, do you think QLB is better suited for this than AVAMAT7?

[Boulder]

I took the QLB matrix from the matrix folder included in HC. It looks like hank315 used the older matrix whereas the one I posted earlier is the more recent one.

What comes to those two matrices, I can't say as I've never used AVAMAT7. The only way to find out is to encode a sample clip with both and compare.

[hank315]

I thought I updated the QLB matrix a while ago, seems I only updated the hardcoded built-in QLB matrix...
Thanks for the tip ;)

[Chainmax]

So, what QLB matrix should I use: the one I posted, the one Boulder quoted or the one Boulder linked to?

By the way, I used both FitCD and PARanoia with the "Resized" setting (who would want to use "Overlayed" anyway?) in order to calculate resizing to 704x480 so as to compare them. The former yielded LanczosResize(656,448).AddBorders(24,16,24,16) and the latter Lanczos4Resize(672,448).Addborders(16,16,16,16). Why the difference?

[Boulder]

The one I quoted first is the "official" one to use since it is a more recent one.

Did you set both programs to ITU- or non-ITU-based resizing?

[Chainmax]

Oh, FitCD didn't have the ITU option checked. Is there any benefit from using ITU compliance? What little I read about seems to imply cropping useful image area and I'd rather have a small AR error than that.
As for the QLB matrix version, the one you gave me a direct link to is different from the one you first quoted, where can I find a direct link to the one you first quoted?

[manono]

Hi-

It's incredibly annoying to fill 128 boxes all the while double-checking you entered the correct number...

Here's how I add a custom matrix to CCE. First I make an ECL using a matrix I typed in sometime in the past. I open it and scroll down to where the matrix is. I cut and paste to replace the old matrix with the one I want to add. I move a bunch of numbers over so they line up properly (you'll see what I mean when you try it), save and close the ECL. I open the modified ECL in CCE. I go to the Matrix window and see my new matrix. It can't be saved like that, BUT, if you change even one number, you can then save it. So, at that point you then have a new CCE matrix, but with one wrong number. Exit CCE, open it again with the ECL which now has one number wrong, or open CCE and just go to that new matrix. Change that one wrong number to the correct number and save it, giving it the same name. OK the request to overwrite.

You now have a new matrix.

No offense to the creator of QLB, but it's so very similar to the Standard Matrix that I don't know why anyone bothers.

[Chainmax]

Still much too convoluted for what should be a straightforward process, don't you think?. In any case, do you feel AVAMAT7 would be better suited to Half-D1, ~1350kbps MPEG2 encodes?

[manono]

Hi-

You can add the matrices in the registry, but I never figured out how. But I also was discouraged when typing 128 numbers as I don't type well, make all kinds of mistakes, and you can go dizzy double-checking all that stuff. I get my matrices from commercial DVDs via DGIndex, so it's an easy matter to get them ready for adding to the list of CCE matrices, and then adding them using the method I described. Convoluted? I recently lost a hard drive and had to reinstall CCE. Believe me, doing it the way I described for the 7 custom matrices I use regularly was a heck of a lot easier than the way I did it the first time, typing them all out.

Half D1? I don't use it. Too blurry for me. I'd rather let the matrix do the filtering. I usually use the Very Low Bitrate Matrix of CCE for low bitrate stuff. But I don't usually let the bitrate get that low either. When doing low bitrate stuff, like DVD extras, I usually do a One Pass VBR (OPV) encode and let the size come out to whatever it has to be for the quality I choose. It sounds like we may be encoding different kinds of things, so if it works for you, that's all that counts.

[Chainmax]

Well, I always try to encode at 8400kbps, but on this special case it's a mix of convenience (having all eps on one DVD) and curiosity (seeing how far can MPEG2 be pushed).

[smok3]

Chainmax: What were some good ("the best") results so far in your tests? (what matrix for hc?)

[Awatef]

No matter the matrix, for a 1.3mbps encode, I would definitely use VCD resolution, of course with multipass VBR MPEG-2.
The results should be excellent (except sharpness of course, something you unconsciously ignore after a few minutes of watching, unlike artefacts showing up unexpectedly on your screen and ruining your viewing experience!)