Which is better? CCE 1 pass VBR or multi-pass VBR

[Gampy]

Hello friends,

I would like to know how to obtain the highest quality possible using DVD2SVCD and the Cinema Craft Encoder.

I'm using the older 2.5 sp version of CCE. I would like to get my SVCD quality as close to the original DVD as possible.

Would 1 pass VBR, using a Q. factor of 2 or 3, yield better results than 3 - 4 pass VBR encoding? Size is not really a concern, since I use the highest bitrates possible with multi-pass encoding. I don't like to have more than 50 minutes of content on a 700 meg CD-R thus most of my encodes require 3 disks.

Professional advice will be greatly appreciated. Informed opinions are very welcome, as well.

ps.

I would also love to get opinions, from experienced SVCD encoders, who have performed side-by-side comparisons of CCE High quality 1 pass VBR encodes against Highest quality 2-pass TmpegENC encodes. Are there any takers on this subject? I'd love to read others' insights on this. :-)

Thanks very much. :-)

[telemike]

My experience with 1-pass VBR is quite good. I usually use a Q factor of 5-15. Quality looks good to me.

[Holomatrix]

Short answer - CCE 3-4 pass

Long answer - I always go with a 3-4 pass CCE encode. CCE is a fast encoder and I feel it is 'too' fast and that it doesn't take enough time to allocate bitrate properly because it is so fast, so I feel by letting it do more passes, gives CCE a change to get the bitrate allocation more "Right" on every pass. Just a side note, I feel there is reason why TMPEG is slower because it takes it's time to get it right the first time, but by doing a 3-4 pass CCE encode, CCE should have it down pat and prevail in the end.

Just my 2cents :)

[Gampy]

Hello Telemike & HoloMatrix,

Thank-you both for your input. You both expressed great opinions! I hope to hear from lots of others, so I can get a good general 'concensus'. :-)

[tylo]

The speed difference of CCE and TMPGEnc is more related to the fact that CCE uses a faster implementation (assembly code) and possible a faster, but at least as good algorithm than the one TMPGEnc uses.

- 1-pass multipass gives about the same quality as one-pass VBR (OPV) with equal file size (kbps). It takes twice the time, but you get predicable file size (=filled CDs)
- 3-pass VBR gives slightly better quality, but takes 4 times longer than OPV. Personally, I don't think it is worth it, even on a fast PC.

Try my plugin that encodes OPV with predictable file size. Set prefered Q to the lowest quality that you tolerate (25 is quite good), and it will figure out how many full CDs it needs.

Download at: http://home.no.net/tylo/

[Holomatrix]

@tylo - I'm going to try your program again but just need clarity on weather or not in dvd2svcd you check 'create vaf' (virtual two pass) or just uncheck create vaf and just leave it at 'one pass vbr' only? (not create vaf)
Thanks

[tylo]

D2Sroba will automatically start DVD2SVCD in 'one pass vbr' mode if you selected '1-pass vbr RoBa'. It does not use the .vaf file for anything in this mode, so you can uncheck it (but it doesn't matter). Please, post your experiences in the ROBA thread.

[Holomatrix]

Ok, thanks. I posted a reply to the other thread because I ran into a problem. here - http://forum.doom9.org/showthread.php?s=&threadid=54570

[DDogg]

Just as a point of information, I did two encodes yesterday. Using a source 1:50 long I did a one pass vbr encode using Q10 for a two disk encode. I used Deen() to increase the compression a bit so that I could get better utilization of bitrate. I also did a tmpg CQ encode using a CQ of 89.5 with the same source and script. Both were very good and I could not tell any difference. Perhaps the CCE 1 pass was a hair better in motion scenes, but could not swear on that at all as viewing was just done on my PC monitor. Someone else with a big-screen might have another opinion (I would sure like to hear it).

Tylos plugin is now a very viable solution and reduces an encode to just a few hours. Given that the plugin will work fine from a Dvd2svcd restart, it becomes very easy and much less time consuming to do a CCE 1pass vbr as a second encode after you have done a standard multi-pass encode in CCE. This will allow everybody to do a easy comparison without it taking forever. You could also throw in a TMPG CQ encode done from a restart.

Users may also want to experiment using undot() and Deen() before the resizing line in the script. Using both only costs me .3 in speed but allows a substantial reduction in Q or an increase in CQ without seeming to effect the encode visually (on my old beater 32 in JVC).

[r6d2]

Originally posted by DDogg
Users may also want to experiment using undot() and Deen() before the resizing line in the script. Using both only costs me .3 in speed but allows a substantial reduction in Q or an increase in CQ without seeming to effect the encode visually (on my old beater 32 in JVC). [/B]
Do you use the default parameters in Deen()? Aren't they too high or just good to keep the visual perceptible quality?

[DDogg]

I just use deen() with no parameters, perhaps you will find a better setting and share it with us.

[Holomatrix]

Why use Deen and Undot before resize? and do you find that this combo combats the DCT/Macroblock problems on solid balck areas? or should I still use Blockbuster(dither) to blend the blocks?

[DDogg]

I was speaking of using those filters to increase the compression without appreciably effecting the video quality. Just for the hell of it, try:

Run a 1pass VBR CCE encode on an existing script. Note size of finished mpv. Then insert undot() and Deen() before resizing (or after if you prefer) and run the same encode again. Note size of mpv [It should be quite a bit smaller]. Now lower the Q about three points and run the encode again. You will find you will be able to lower the Q by 3 to 5 points and achieve the same sized mpv as you had in the first test encode that did not have the filters. If you want to do this quick, add selectrangeevery(1500,15) to the bottom of the script.

Whether that works for you is strictly for you to say. This works for me, given my outdated viewing method, without effecting the video quality [much] (my subjective judgment) and allows me to get my 1 pass q range down less than 10-12 on most of the under two hour vids. That is a quality level I am happy with. Please do understand I am not real particular about quality anymore so long as I don't see anything obvious or distracting. I just like to watch the movies.


Addition: Interesting. I was just using Tylo's 1 pass VBR plugin and doing a 4:3. I wanted it on 2 cds, but it was going to take a Q34 to make it fit with a standard script. So, I threw undot() and Deen()into the script and it now looks like that will allow a Q of 22. That is 12 full points of Q reduction!

[markrb]

Will any of these filters be of use in a DVD -> DVD conversion when no resize method is being used?

Typically an average bitrate can be as high as 4000k or higher for this.

As always just looking to play with new toys.

Normally I would just try it and give the results, but right now restraints on my time have forced me to ask.

Mark

[lonewolf454]

I have been using the following settings since last year for all movies and have had absolutely marvelous results, playback on Pioneer DVD 503. CPU Pentium 4 2.0 GHz, 512MB DDRAM, average total time for dvd2svcd to complete is 5-6 hours, per 1.5-2 hr movie.

NTSC Field Operation = Automatic,
iDCT Algorithm = 32 Bit SSE MMX,
Audio 192 (Unless movie is real short, then 224),
Autodetect Azid = checked,
Multipass VBR = 5,
Image Quality Priority = 20 (Used to change for hi action, but 20 seems optimal),
Using VCDXBuild, DVD Chapters, PBC Use Selections,
Include XML on cd, Use MPEGAV,

Always check the ripped files from the dvd with bitrate viewer and see what Intra DC is (8-10), Progressive or Interlaced, ZigZag, Linear or Nonlinear, and set the advanced settings under Encoder tab accordingly, especially the Intra DC. I always check the vobs with bitrate viewer before encoding. If all vobs show interlaced, then do smartdeinterlace.

I check the total movie length and set for 2 cds under bitrate tab. If movie is 100 min, set 1 cd for 50, 2cd for 100.

I then load the file with Daemontools, check it first, then burn it with nero copycd.

Piece of cake and perfect results. Believe me, I have had hell and terrible results in past until I started doing this method. If better result is obtainable on a cdr, I dont know but these look great on my 32" Trinitron. O yeah, I recommend buying quality cdr, made in Japan, I bought some cheap cdr on the net last year and they already have deteriorated. CHeck on the label for made in Japan, not China or Taiwan!

Good luck.

[DDogg]

Normally I would just try it and give the results, but right now restraints on my time have forced me to ask. I don't do DVDs, but with the selectrangeevery(1500,15) added it should only take you 3 or 4 minutes to evaluate the potential size reduction as a projected percentage of the full encode.

[davidh44]

I'm a D2SRoBa user (great plugin!) and will be trying the undot and deen filters being discussed in this thread soon. Here are some comments by the Undot author (trbarry) that I found in various posts that may be of use in deciding where to insert these filters:

Most Avisynth filters work on progressive frames. UnDot is no exception. So it (and most filters) should be after the Telecide or deinterlace step.

Some folks have found even better compressibility doing it both before and after the resize (undot is cheap on processing power), but I have not tested this myself. Assuming you are downsizing it is a slightly stronger (not necessarily better) filter after the resize. But I have usually tended to put it before resize, for no particular reason.

And my own feeling for it is maybe to have it precede smoothing type filters like Deen, C3D, UnFilter, etc. But I don't now if that is important.