cvd far worse than svcd !

[waldok]

Hi,

considering th hype around cvd, half D1 and blah blah, I decided to give it a try. Using the same ecoding parameters (bitrate, quality, vbr) I produced both a svcd and a cvd with dvd2svcd from the same chapter of the movie "the others (chap 12 if anybody is interested).
Results : cvd quality is not acceptable on my 109cm toshiba tv set. you can clearly see aliasing there, particularly on "nearly-vertical" diagonals. The only reason I can see is the 352 pixels width, which, once interpolated to the TV, generates this very visible stairs effect.

SVcd with the same parameters looks great when compared to cvd. can anybody explain me if I did something wrong ? Is this all I can expect from cvd ? Then I really don't understand all the hype about it...even if it is directly DVD compatible as a MPEG stream...quality is not ok.

advices and help appreciated.

Waldok

[mrbass]

I suppose I'll have to try it out myself. But I believe I won't like what I see. This kinda reminds me of using s-video with my dreamcast. It showed Dead or Alive 2 clearer and it looked better, however, you could see the rough edges of the characters. In other words, the developers counted on the fact that regular tv connection would be the blur/smooth effect and people wouldn't notice.

I suppose if you have a cheap tv and cheap dvd player then perhaps you won't notice cvd compared to svcd. I have a pioneer 434 with sony wega 32" tv hooked via component cables and I believe I will notice a difference between cvd and svcd. Will try to do a test soon.

And yes I will give a higher avg bitrate to cvd since that's the advantage of it. I think you really need to do a full movie because a single chapter is going to give high bitrates (the same) to both svcd and cvd.

BTW if you really wish to test out encoding action I suggest "Ultimate Fights: From the Movies". CCE needs at least 2200 to look decent for NTSC fullscreen. I did cbr and vbr 4 pass, bicubic and I'll look at the results more closely. But it just reconfirms the notion that bitrate above 2200..vbr is a waste of time and it's same as using CBR. Cuts your encoding time by 80% too (by 4/5) I tried putting this on one cd (53mins) but way too may artifacts.

[waldok]

@BAch : Did you mean bicubic or bilinear in your first answer, cause you used both in the same sentence. I tried it with bicubic 0, 0.75 and didn't add borders, encoded in 4:3 since I have this 16:9 switch on my TV set (european model). So I'll get it full screen (apart from 2.35 ones when I get a new widescreen set.
Your information about compared quality is interesting in your second post but I doubt I'd be satisfied with a bitrate between 1000 and 1800. So I may not be a good client for CVD...

@MrBass : I did the whole movie in CVD and I was very disappointed, that's why I started experimenting with only one chapter afterwards. "The Others" is a quite dark movie since in the plot children are not allowed to be exposed to sunlight. Couls this be the cause of my problems ? My guess is this may produce macroblocks artefacts, but probably no aliasing effect.

In fact, I can't imagine how playing around with the bitrate could change anything to aliasing, since it generally is due to poor interpolation of missing rows of pixels. Does the interpolation from 352 to 720 (or is it 704) occurs at display time or at decoding time ? In the first case, there is probably no way to correct this in my humble opinion (never seen a TV set with a "column doubler"...Experts opinions are welcome.

Last, but not least, I finally gave the 720x480 resolution a try in DVD2SVCD. This is no standard SVCD either, but my TOkai (aka Yamakawa or Raite) plays it fine and I can tell you the quality rivals the original DVD : no aliasing (of course not), sharp picture, no macroblocks artifacts, the ideal solution. I'm currently running the whole movie encoding this way, I'll let you know what it gives and how it fits on 2 or 3 cds. I tried this on one chapter and it was absolutely amazing. PLease DVD experts, tell me if this resolution can be directly used on DVD-R like CVD does (must be full D1 if I'm not wrong) ?

Sorry for long post, I hope we are starting a very interesting discussion here.

Waldok.

[SatStorm]

For the NTSC users, CVD ain't a good option most of the times. CVD is a PAL thing. You are lucky if your R1 standalone support the format and more lucky if it plays it correct! Also CVD is not for Newbies and basicly, it is simply a fair alternative IF you want something beyond VCD/SVCD/DVD. xSVCDs gonna look always better any CVD.

A tip, sometimes help: If you encode with TMPGenc, don't use mjpeg codecs while you grabbing, only Huffyuv. Also, use the sharpen edge filter, it helps the sharpness a lot, but don't go beyond 32,this filter add noice!

With CCE you don't have those problems. Use whatever you want. This is another reason why CCE is better (and much more expensive)TMPGenc.

About picture quality if you compare CVD vs SVCD, in my sharp LCD videoprojector, with component connections and a 2.5 meter screen as a target, I see no difference between them. The same also with my Nokia 32'' tv and with a 36'' Philips Pixel plus TV I just tested.

My source is interlace PAL DVB transmissions and my output is intelace PAL VBR CVDs always.
Anyway, if CVD ain't good for you, don't use it!
Some people are happy with it and some don't!
Personally, I like CVD more SVCD, that's why I publish those infos on VCDhelp.com. It is simply another alternative for our hobby!

have fun!

[waldok]

Sure, I never meant that CVD was bad. I just meant that comparing both SVCD and CVD, my preference goes to SVCD. Of course, since I don't like it that much, I won't use it anymore for DVD rips. My point here was just to know if maybe I had made a mistake in settings or something that would explain the "not so good" quality.

ANyway, I think I'll keep up with my XSVCD (720x480NTSC and 720x576PAL) since it gives astonishing results.

Maybe I'll give CVD a try for PAL to see if it is better.

THanks for your answers

Waldok.

[mfshva]

at first Sorry for my bad english.

I think with CVD and SVCD it is like Longplay (LP) and Shortplay (SP) at VHS Tapes.
CVD is like LP i can burn a little more Movietime on CD-R without seeing many QualityDifferences on TV to a SVCD. And i can burn it later on DVD-R

If you looking only for Quality SVCD is your Way.
If you not need the highest Quality but will burn your Movies later to DVD-R and backup many LongMovies (2Hours and more on 2 CD-R´s) then CVD is your Way...

So anybody must set his own must have. For me SVCD is a little bit more Quality as CVD but on CVD i can burn more Movie and it is invissible on my TV that it is going Bad Quality....
And CVD is faster CCE SVCD (bicubic Realtime: 0.5-0.6)
CVD CCE (bicubic Realtime 0.8-0.9) on AMD 800 Duron 128MB Ram

[waldok]

Thnaks a lot for all the advices here. I'll play aroun dwith the settings a bit.

Waldok.

[r6d2]

Originally posted by Bach
Let me try to make this CVD hype more clear:

=>quality comparison:
->paramethers:
average bitrate (avr);
bitrate/block (bpb);
visual quality(q).

Results:
a) for bpb=constant:
q of DVD > q of SVCD > q of CVD > q of VCD

b) for avr=constant: This is not so static but, most of times, we have the following:

b.1) if avr > 3500 kbps:
q of DVD > q of SVCD > q of CVD > q of VCD

b.2) if 3500 > avr > 1800:
q of SVCD > q of CVD > q of VCD > q of DVD

b.3) if 1800 > avr > 1000:
q of CVD > q of VCD > q of SVCD > q of DVD

b.4) if 1000 > avr > 600:
q of VCD > q of CVD > q of SVCD > q of DVD

if avr<600, any standard resolution looks bad with mpeg2. Maybe we should try 64x64 resolution:)

So, it is not simply a "what looks better" question.

@Bach,

This is a very interesting comment. Did you come up whit it by trial and error or there is "math" behind it?

Would you please explain the math behind it, if any? I am really interested in this matter for a sort of AI model for a "Movie fan delight" guide I'm working on.

Many thanks in advance.

[Kika]

The Key to clear and sharp CVDs is Resizing - and Cropping! Bicubic is not the right choice for this Format. Bilinear will give you better results - if the cropping is done the right way. Never resize to a size which is not devidable by 2. 704x480 -> 352x480, that's the ONLY true way.

[Kika]

Nonsence :rolleyes:

I have never seen any DVD source with 704x480 resolution...

I have some of them...! And also PAL-DVDs with Picture-Size of 704x576.

. If you achieve such resolution using crop, most of time you will remove some of the picture. Great deal

720x480 is using an overscan of 16 Pixels. The Motion Area of EVERY DVD is 704 not 720. Guess you have to read the ITU-R-Specs.

there is no rule "your resize must be mod2". Prove your theory.

It's proven since years. Ever heart about the "Nyquist-Effect"?

btw: you are the author of those "make me a moron" matrices?

? No idea what you mean.

[Kika]

Calm down Bach... :rolleyes:

It seems that those specs are nothing but bullshi*

Come on, you realy think you are right and the Specs are wrong?
Do you know People like TheWEF or mb1? They are also spouding Bullshi*? Don't be silly.

misconception

Oh, these side. Yes i know it. But think on what we are talking about here, a resizing in one direction.
And the Nyquist Theorem is a fact. A lot of people (mb1 also) have done a lot of resizing tests to prove this.

"kika trick film"?

OK, talk to me. What's wrong in your oppinion? Do a factual criticism. Talking about "morons" is a "little" emotional...

[dvd2svcd]

http://www.uwasa.fi/~f76998/video/conversion/

[Kika]

Or this one:

http://toolbox.sgi.com/TasteOfDT/documents/video/lurker/pixelaspect.html

[gerti67]

Originally posted by Bach
nope. It is not my only opinion. Given the lack of pattern for the DVDs sold here in Brasil - which are mainly encoded at USA - those specs are bullshi* in the opinion of professional encoders also. Do you want a sample of (for example) "cast away" to realize it? I don't care what mb1, TheWEF, Jukka or anyone else think. The DVDs are in the stores to show they are wrong.Oh, come on Bach, really, the ITU standards are the base for all digital/analog video technical equipment and so further and so on. And the fact that some freaking id*ots don't do their homework and author something that merely deserves the label "DVD" on it is not the fault of the specs nor does it mean they're wrong either.
Originally posted by Bach
Nope. You do not know it. Read it seriously and don't come here with "we are talking about a special issue". I will give you a quote from that site. It goes after a good math explanation and real world test: "My conclusion is that the bilinear method is preferable for small deviations in image size while the bicubic method is better for large deviations in image size." Well... from DVD to CVD we have a resizing of ~100%. I guess it is a deviation large enough:D
...
The Nyquist Theorem is a theorem. Nothing but that. ...In my opinion this theorem has something to do with it because minimal resizing tends to create finer patterns which mean higher frequencies for the MPEG encoding step and so this effect kicks right in because higher frequencies need higher "scanning frequencies" according to the theorem which for the MPEG encoding step would mean matrices that doesn't limit the "scanning frequency" too much and what counts even more is the analog world - when watching your CVD/SVCDs on your TV - with the "improper" resized movies your TV can exhibit blurring and moirée effects because the sampling frequency of the TV is too low and therefor can't properly recreate the whole details of the picture - exactly what the Nyquist theorem postulates - in more or less rough words. And, although I did not test it (you can blame me for that ;)) - but only guessing by using common sense tells me: Resizing a picture mod 2 aware can sure help the resize algo not to create even more higher frequencies.

Just my 2 cents,
Gerti

[Kika]

:devil: Hehe, that's very funny, like it...:D

Bach, read you own Postings, than mine, think 2 Minutes and all will be fine, belive me... ;)

It was Kika who said that the resizing should be at mod2. The sentence is correct.

So where's your Problem?

He was saying that you should resize from 704 to 352. So, instead of to say "Never resize to a size which is not devidable by 2", he should have said "Never resize using a ratio which is not equal to 2." This is the refused sentence.

OK, that's a point. My english isn't good, sometimes my writings may be a little clummsy. But you've also done a "strange" posting.
"I have never seen any DVD source with 704x480 resolution. If you achieve such resolution using crop, most of time you will remove some of the picture. Great deal..."
That was your posting. And later then:
"However, almost 90% have some value between 720x480 and 690x360. I will post a sample for you if you want"

That's imposible. I have never seen a DVD with a Resolution of 690x360. I have seen such Discs with a Pixel Area like that, but that's not the resolution.
Guess now it's 1:1 in Mistakes, eh? :cool:
The 704 -> 352 i've posted was meant as an example. And, to say it absoluty clear: It's always better to resize at mod2, no matter what type of resizing you use.

About the Matrix: There are two Versions of it, don't know which one you have testet. The older version is OK, but correspondents to a Bug in CCE. This darn Encoder don't like an 8 as first Element of a Non-Intra Matrix. No Idea why, MPEG alows the 8, and with TMPGEnc, there's no problem. Later, for use with CCE, if changed it into 16, but that's not the best thing.
The Matrix was created on TMPGEnc and works on both encoders, but works better with TMPGEnc. I Don't know how familiar you are to DCT and Quantisation. There IS a difference between the CCE CG-Matrix and my creation. The Matrix was created for getting less Quantization-Errors, how good it works, depants on the Bitrate you use. And it only works on Painted Grafics, not CG-Animations.

Remember that the "high frequences" of a picture are the edges, and if at 720x480 an edge ...

Hum, Frequences have nothing to do with Picture Sizes. If you are talking about Frequencies, than you are talking about the DCT and than you are talking about Blocks of 8x8 Pixels.

The Nyquist Theorem hast to to with them all: Resizing and Quantisation.

If you are talking about Nyquist and resizing, than it's a rule for what you should do and what not (mod2 and other things).
If you are talking about Nyquist and Quantisation, than we are talking about, how to set the elements of a Matrix. But at this point, i must pass. Can't explain that in english, sorry (hey, that's complicated even in my native Language... ;) ).

Regards,
Kika

[r6d2]

Gentlemen,

I'm sorry that a simple question I asked Bach if he would kindly respond has become a flame war on specs and resizing matters.

This thread was sort of dead, anyway.

I guess I will have to look for the answer in some other place.

Thanks anyway.

Regards,

[waldok]

I like it when a thread of mine, nearly 1 year old, is brought back to life for a good flaming party :D

So coooool...

Waldok:cool:

[Kika]

@Bach

Did'nt you read what i wrote? You have quotet it...

Frequences have nothing to do with Picture Sizes

I've not talked about resizing there. I talked about your not clear and not correct using of terms. But i see, our Points of View are getting closer...

And BTW: High frequencies are not only the edges of Objects. Interpreted as high Frequencies are also the Motion Compensated Parts of B- and P-Frames.
To your Exampel: Nothing to say about that, but we talked about other things. However, the Example is a good explanation of what happens to a resized Picture while encoding to MPEG.
Now all we need is the same Example before and after the DCT with differend sizes an Matrices. Then we have a really good Guide about DCT. :)

Maybe your Picture with 704x576 (or, if you like it more, 720x576) resized to 352x576 or 480x576. Displayed as a Matrix before and after DCT.


About Motion Areas and other things. It was not my idea, all i posted was about the Specs. I have lot's of DVDs which are following the Specs completly, and i have other one's with... hum, nearly crazy..., settings.
Oh, and the Pixel Area (or Image Area) must not be the same as the Motion Area, it should be, but must not.