KVCD: Can I still use DVD2SVCD?
 

This will be a fairly long post but I'd far prefer to put all the questions in one spot. Yes, I've read the guide and played with ToK a little bit. I'm confused and looking for specific, guided help. I'm NOT working with .vob files, only AVI of DV camcorder or analog captures. I'm not new to video editing and compression, just new to KVCD and this board. Most, if not all of these questions are probably answered in various other places. I've been looking through the boards for a couple of days and I'm quite confused. Hence, this post.

I'm using DVD2SVCD to convert my AVI source files to 1/2 D1, 2/3 D1 or D1. Usually, I do all my filtering with AviSynth and feed the encoder the processed file and the desired min, max, av video bitrates, audio rate and frame size. Audio is always left at 48 for migration to DVDR. CCE is my encoder of choice because of compression time.

The only exception to this is quick encodes for which I'll edit the AviSynth script when DVD2SVCD is done with audio transcription to force an overlay frame number.

Most encodes have a target size or around 770M because that will allow an easy migration to DVDR. (6 CDs = 1 DVDR with menus.) For high motion, I've been encoding MPEG2 with ave bitrate of around 5500 and max of 6500.

If at all possible, I'd prefer to stick with DVD2SVCD because I'm familiar with it. If there's a compelling technical reason, I'll switch but would prefer not to do so.

Another type of encoding I do is 720x480 MPEG1 with TMPGEnc. Filesize is not as critical with these encodes because they are short and used inside PowerPoint presentations. TMPGEnc easily allows 15 fps which has shown to be important for use on a wide range of office computers but the KVCD encodes just might work at 30 fps as they appear to dramatically lower bandwidth. If that's the case, I'd love to use CCE.

My understanding, based on a kwag post at doom9, is CCE can use the matrix and q values for kvcd. I've also noticed things have changed with how the templates are provided here in the past couple of months.

OK, here are the questions:

Can I still use DVD2SVCD or must I switch to automation tools posted here?

If I must switch, which options give the highest amounts of automation and accuracy, preferably allowing manual editing of the AviSynth script?

If DVD2SVCD can be used, which of the various size prediction calculators have shown to give highly accurate results?

Do any of the automation tools located here support batch processing of a series of source AVI files?

wrt CCE, how is it properly configured for the custom templates?

wrt TMPGEnc, same question.

I saw mention of a suggested maximum bitrate of 3000. Is there a problem with higher bitrates?

wrt keyframes: some of the stuff I make is MPEG1 and needs to be as portable as possible. The drawback I've seen with "regular" MPEG1 files is slow response with the position slider in Windows Media Player. Would the KVCD templates provide better performance, either through less bandwidth or allowing more frequent i-frames in the same space?

Good point, Kwag, about the mini-discs. VERY good point.

Re: A number of basic questions
 

Hi Fred and welcome :D,

You can use DVD2SVCD, but I believe ToK's file size prediction is more accurate. I'm not sure if DVD2SVCD has updated his file prediction algorithm, but ours is very accurate.The best quality will be achieved with the optimal script posted and updated here: http://www.kvcd.net/forum/viewtopic.php?t=3483 Because we are using CQ mode for encoding, you can only use either DVD2SVCD which has prediction, or you can use ToK. Mind you, DVD2SVCD already has included the KVCD Q. Matrix. It's already built in. But I'm not sure it is the latest "Notch" matrix :roll: :twisted: No. hopefully some day :!: You have to patch the matrix into CCE with a program called CCEPatcher.Specs here: http://www.kvcd.net/e107/article.php?21.255
Just open any of the TMPGEnc templates, modify the matrix with ours, and save your template with a new name. This way you can customize your own templates with KVCD parameters ;) Not at all :!:, but only if your target (DVD player, etc. ) supports it. You get the best quality per bandwidth with MPEG-1 and KVCD parameters. If you want a better player, one that behaves better than WMP specially when movind the slider, please check VideoLan Client Media Player: http://www.videolan.org/vlc/ ;)

Also, remember that although you can use CCE for MPEG-1, the quality will never approach TMPGEnc's MPEG-1 quality. For low bitrates, CCE and TMPEG are about the same in MPEG-2. For higher bitrates >4,000Kbps, CCE is a better encoder. And I believe that MainConcept is even better than CCE in MPEG-2 for high bitrates :!: You can also try Canopus ProCoder. I haven't played around with than one too much.

-kwag

Just looked at the script. OK, this is for low to medium motion stuff, right? Looks like high-motion would get smeared.

Is the benefit of your matrix directly related to the smoothing script?

Would the calculators be "busted" if I don't use the script? (edit: ignore that, I just noticed the linear adaptive aspect)

Would LOVE to use another player ap but you'd be surprised how confused some people get just using an HTML-based interface. VideoLAN is quite nice, been following that for quite a while. If I ever netowkr a house and add a video server...

That's exactly the idea :!:
You can't see details on high motion anyway, so we blur to remove sharp edges, and the encoder will compress the data much more than with sharp edges moving around. It's suitable for any material. Not at all. The matrix is optimized for low bitrates, and specially optimized to lower quantization on the low frequency spectrum. That's why the "Notch" name is in there :) It notches out the low frequency (dark/low lit) scenes which normally show up as visible DCT blocks on dark scenes.
So you get less visible DCT blocks with KVCD's "Notch" matrix ;)You can use the standalone VC media player executable :idea:

-kwag

ok, I see. Sounds like this would have an effect similar to what I've been asking at doom9 about an adaptive smoother for swim in the near-black static areas.

I'll have to give it a shot later today. The vast majority of my stuff is high-motion with high bitrates but this will still help. Wonder how it will do with flash photography. That usually beats the snot out of most smoothing.

Looks like the scripts are for progressive frames. I'm currently running STMedianFilter on each field then recombining. Have you tested your scripts that way with MPEG2 compression?

You know, maybe the player client can be launched from the CD. I just checked the VideoLAN site and it sure appears that way. They've done a complete redesign over there. Will look into that later. Thank you very much for the tip. Maybe it could run from the CD and file be dgbob'd with your encoding for MPEG1. Man, that would be great.

We have tested STMedian up and down, and the temporal filtering part is VERY bad :!:
That's why we use only the "spatial" part of the filter, and use TemporalSoften for "temporal" filtering ;)
The adaptive script should work just as good on either MPEG-1 or MPEG-2.

-kwag

ok, thanks for the insight. You've probably done far more exhaustive tests than I have.

Actually it wasn't my findings 8O
I just got many reports on STMedian filter from many people, and I decided to try it out, and they were right :)
That's why I settled for STMedianFilter(8, 32, 0, 0 ) which applies spatial filtering, but zero temporal filtering.

-kwag

I'm rapidly starting to think your script with a few additional thing I need like a dominance-sensing field swap (DV) just might be what I'll use for all cleanup. Still need to torture test it and modify to be field-based, not frame-based for source that is to stay interlaced but my hunch is the cleaning/smoothing part is highly optimized.

One of the things I'd asked about lately on doom9 was enhanced smoothing of near-blacks in static screen areas. There was a very interesting reply regarding masking. That's something I want to examine about your script. Swimming near-blacks can eat up a lot of bandwidth.

I also noticed undot. Guess that's to get rid of specs, huh? Just registered SpotRemover for VirtualDub to do that. Apparently, its internal routines are YUY and I've asked about a way to send that instead of RGB from AviSynth.

The script you have looks very nice.

Kwag,
so, does main concept able to make use of your matrix. does main concept able to put 2 movies onto 1 DVD-/+R?

You will, but the quality will not be very good because the bitrate distribution is not very good. If you use an average bitrate that is much lower than the max, MCE will never reach that max, not even on high activity scenes. So you'll have a very blocky picture on action scenes.

-kwag

Am I correct that ToK and the KVCD-specific calculators are focussed on estimating the performance of CQ compression with TMPGEnc?

Yes. ToK finds the correct CQ for a given file size target :)

-kwag

OK, so, when a goal is low bitrate use TMPGEnc and TOK if filesize is very important. As the bitrate gets higher, the importance of TMPGEnc lessens because CCE's performance starts to catch up and its speed is a benefit compared to TMPGEnc. Both should be used with your improved matrix and Q settings as well as proper filtering prep.

To use your mini-CD example, a credit card CD might have, say, 20G available and ToK would help to optimize that space.

Not exactly! You can use ToK for any target, even for DVD movie size prediction. This way we can fit ~6 hours Full D-1 resolution with KDVD and ~10 hours at Half D-1 :) At around ~3,000+Kbps you can use CCE but ONLY creating MPEG-2, because CCE is not very good (never has!) on MPEG-1 compared to TMPEG :!: Yes.20G 8O :?:

-kwag

Ah, yes, well, um....20GB in a credit card CDR would be something, wouldn't it? I just checked at Meritline and they're 50M (I was thinking 20M.) Until learning about these techniques, I thought that was too small to do much of anything.