SVCD Samples -- How they were created?

[tandr]

Hi folks,

I downloaded these 2 SVCD 2 samples from SVCD page, and they look... well, nothing short of amazing. Their quality inspired me to get my hand in this, and I am trying to convert my home shootings to svcd (using tmpgenc). But, unfortunally, the quality that I get is far from quality of these samples even with much higher bitrates.

So, question is... how they were created -- what tools, what settings, how long did it take etc?

Any info is greatlly appretiated.
Thanks,
tandr

[DJRumpy]

You have to remember that these samples were probably created from a digital source that was exceptionally clean, like DVD. If your trying to convert a CAM source, or a broadcast capture, those are two entirely different animals. Cam sources have alot of motion, unless a tripod was used. This "shakey-cam" motion makes mpeg encoding difficult (think blair witch project). Broadcast material is just as bad, due to the 'noise' in the analog signal. What type of source are you trying to convert?

For a decent comparison, do a small dvd rip, and convert a sample from that. You'll get simular results.

[tandr]

What type of source are you trying to convert?

grabbed from my DV cam to the disk as 720x480 NTSC AVI file

I dont think shaky cam is biggest problem there. May be it is that picture is "noisy" somehow (although I dont see any noise, but some pattern should be there, because if I move camera over clear sky, I can see some minor "mesh" of more bright pixels ). May be I need to apply some "denoiser" filter.

So, question is remains -- what soft and setings were used for the samples. Just out of curiosity

thanks,
t

PS

This "shakey-cam" motion makes mpeg encoding difficult (think blair witch project).

heh, what you are saying that "Blair witch" DVD was hard to rip and reencode ?

[deh707]

[/quote]heh, what you are saying that "Blair witch" DVD was hard to rip and reencode ?

well, the blair witch movie itself, the camera guy had the cam shaking like big time throughout the movie, am i right?

[deh707]

wait a minute, how do u use "Quotes"???????????? im a noob to forums

[vhelp]

Hi guys.

@ tandr,

>> heh, what you are saying that "Blair witch" DVD was hard to rip and
>> reencode ?

no, DJRumpy was saying that the source was ShAkEy in some scenes, giving
you the impression of "home-footage" from an ameture.

Some things to note, when shooting home footage w/ DV cams..
* Light source. (though, I don't understand you issue w/ blue-sky)
* movements (blare-witch project)
* stabalizers (see notes)
* Interlace.. mother of all evils (for dvd reproduction)

Digital Stabalizers (DS)..
* those cams w/ these DS don't mean that they take out every bit of shake.
In fact, they don't!! What they do is take out those tiny vibrations in your
hands, when you are holding the dam. Have you ever seen an older person
w/ the shaky hands. And, have you ever seen how there fingers sort of
vibrate ever so small, unitl you look long and closely enough at it to notice
the shakes ?? Well, that's basically what the cam is detecting and count-balancing
for. But, if you shake or jerk the cam, there's no stabalizer in your cam to
reduce this. So, factor that into your encoding issues.

Interlace..
* Mother of all evils - and is true. But, for straight TV viewing, from your
DV cam and/or miniDV recorded to, tapes - to TV viewing, the quality is
perfecto. That's because there is no Encoding (or re-encoding) to say,
MPEG-2 and w/ Interlace (which makes things SOO much worse) in the
final encoding, then played and watched on TV sets.
As I was saying.. There is no form of reproduction of the source. The source
is connected to a TV set and when played back, if perfect. BUT, what if you
transfered it to your harddrive via firewire ?? Well, if you were able to play
this to your TV set - - Guess what ?? it would be perfect too. Why ?? ..dahh !!
Because there was no re-encoding.

Now.. lets talk some more about Interlace for a moment.
* When you encode Interlace source, you will undoubtfully get poor quality
in your final MPEG-2 encoding. But, if you use high enough bitrate, then
your finaly MPEG-2 quality shouldn't be too bad at all, if at all. BUT, that all
depends on HOW you handled your CAM shooting, light source, shutter speed,
white balance, focus, exposure and other various controls on the cam, lens,
etc, etc.
Now, all these settings DO effect the "quantity" of Interlacing - - that is, the
level of hardness in Interlacing.

Encoding Interlaced DV CAM footage..
When you encode such sources, and assuming correctly - - to MPEG-2 and
w/ Interlace, your bitrate has to be adjust as high as require, but w/in the
bounderies set forth for your given encoding projects ie, SVCD vs. DVD
If you are encoding to SVCD, expect to have poor quality encodes, and also
add/factor the above issues too.
Now, when encoding to DVD, and w/ DVD specs, the only thing you can
blaim on the final quality is the
1 - Interace
2 - above listed items of issues
3 - and, for the most part, too much swinging/jerking and so on.
4 - Oh, did I say, your DV cam (altogether) too ?? could just be the level
......of the make/model ie, not a professional grade.

Lastely, let not foget other factors as well:
* your level of knowledge in the video field ie, DV cams, MPEG encoding etc.
* your encoding skills !!
* your DVD player
* author app
* CD media
* TV set
* did I forget anything ?

That's about it for me thus far.
-vhelp

[tandr]

vhelp, u puzzled me BIG TIME with your answer. Now I got that interlacing is bad (somehow) but "now what" question is still there.

About blue sky issue.

I guess.... scratch it -- I know that every CCD was not born equal. So, some of the pels are more "sensitive" that other and (AFAIK) that sensitivity mask is quite stable per CCD (camera?). Now, if u move camera over almost solid background, u will see that some kind of mesh made of more brite or bit dark (another color?) points sorta moves with your. You can think about it as of some sort of opposite of "dead" pixels in LCD screen . I dunno if this effect is something known and noticible by other, but for me it is "noise" and I see it on my 57" tv when viewing my footages. I dunno, it could be not CCD issue but some software encoding /comressing algorithms in camera that cause that, but I saw this effect in many footages.

I guess 3 CCD cams less prone to this because
a. they are more expensive -- better CCDs used
b. 3CCD can offset errors of each other and smoth them (I guess)

So, folks, the question is remain. Minus my shaking hands (well, I am not THAT old yet, but have this problem with hands anyway), what should I do in order to get same beautiful SVCD as samples provide ? Deinterlace and smooth it, and then encode?

thanks,
t.

[vhelp]

hi tandr,

sorry if I sounded poowy it wasn't ment to sound that'ish.

I'm puzzled by your blue sky issue. But, I think I know what you are
talking about. Maybe I'll do some explorying w/ my Sony DCR-TRV22
and see for myself what the big deal is. Maybe it was just your bitrate
or who knows.

The old guy thing..
I was just trying to give you an analigy of how the Stabablizer works. I know
it was rough, but do have a hard time explaining myself at times.

de-interlace..
The majority of falks here (and there) claim that it's best to encode your
DV source w/ Interlace on. That is true, but only to a point (IMO)
IMO, I "feel" that de-interlacing is better, but if done very well. And, there
is no perfect de-interlacer around. I've found none.

de-interlace and quality..
You can de-interlace your home-footage, but the quality w/ not resemble
100% of your sources originality. That is, becaues one field has to be descarded,
you can't help but to notice that quality has dropped, though to some people,
they can't tell the difference. That's up to the individual and his/her level of
skills (in a facets of the video process from DV cam to MPEG-2 encoding to
final media authoring)

Interlace..
If you want your video to play smoothly, then definately go w/ Interlace.
The only major problem w/ Interlace encoding is the fact tha once you take
your DV from an .AVI format, and enocde to MPEG-2 (but w/ Interlace) you
end up w/ a lot of degraded or poor quality, though it will play smoothly..
giving you the impression that it's very good quality, which usually works
for most people. So, encoding w/ Interlace is the way to go.

What I said..
I didn't say that you should not (but if I did here or anywhere's else..)
I was saying that encoding to MPEG-2 w/ Interlace will not improve your
source's quality. You will have Interlace artifacts due to the Interlace - - well,
at least during the Interlacing areas. You'll see the blocks/pixelation there.
Just magnify those frames, using vdub 1.4.13 (has MPEG-2 support) and
see for yourself. do a few encoding clips at varous bitrates ie, 2000, 3000,
4000, 5000, through 9.8k and see which one STOPS the artifacts. Then,
you'll know where to set your bitrate to. Use a fast scene though.

best quality..
The only way to minimize the artifacts is to shoot in Progressive mode.
But, the problem is, those kinds of CAMS are expensive $$$$.$$ and the
CAMS of today, are Interlaced. But, if you cam support Progressive, then
definately use it, and encode w/out Interlace.

-vhelp

[tandr]

thanks for explanation, vhelp. I will try to play with different bitrates and filters to get rid of noise. I am using tmpgenc, have no luck in using vdub, cannot get it to read my AVI.

Is it possible to create custom filters for TMPGEnc?

I forgot to place after old guy joke, so dont be hard on yourself

t.

[vhelp]

@ tandr..

actually, vdub does open DV files from your DV cam. You are prabably
doing your DV stuff using Type 1. vdub can't process Type 1 DV files,
and hickups w/ an error (more like a warning) message about the audio,
and ends up leaving OUT the audio.

However, if you do your DV stuff w/ Type 2, vdub will open fine.

>> Is it possible to create custom filters for TMPGEnc?

No, it's not. But, AVIsynth has more than enough filtering to tweak w/ your
encoding. And, if that's not enough, well, as I soon as you clear up your
vdub issues, you'll have those filters to tweak w/ too.

See ya later.. it's lights out for me.
-vhelp

[DJRumpy]

tandr, you can get a DV Converter in the tools section (there are three or so..two are freeware if I recall correctly). You can use these to convert your file to Type-2. This will take care of your VDub issue. Just search for "type" on the tools page.

For you video, I definately agree with VHelp. Leave it interlaced. You should look at encoding in CVD, rather than SVCD. The reduction in resolution gives you more bang for your buck where bitrate is concerned. The lower resolution requires less bitrate to produce, freeing up bitrate to handle image data. You end up with a slightly softer image than SVCD, but the quality can be better. Again, as vhelp said, if you can do DVD, than that is the media of choice, for cam sources because of their higher bitrate demands.

Even though your cam video may not appear shakey, if it is a hand held video, then it has hand motion. To understand why this is important, you need to understand a little bit of the MPEG encoding process. It uses a key frame to encode a complete frame. From this frame, the next frame in line is examined. Only objects that are 'new' in the next frame are actaully encoded. Items that have not changed from the key frame are not re-encoded, until the next keyframe (keyframes can have any number of non-keyframes in between them). The encoder also looks for any moving objects (like a bird flying for example). If it properly detects this motion, then it will simply say something like 'object x moved 3 pixels down, and two pixels to the right', rather than re-encoding the bird on the next frame. That saves bitrate for other parts of the picture that need it. This is an extreme over simplification, but it will do for examples sake. Same example, but say your hand is shaking, and the background is bobbing up and down. If the motion detection function in the encoder doesn't properly catch any motion going on, then it has to use bitrate for these 'new' objects like the background. They are considered new, because the encoder didn't catch that they simply moved from one frame to the next. It then uses bitrate to encode that image area in the next frame. Knowing this, you should begin to understand why motion eats up bitrate. Some encoders are better than others at detecting motion.

For TMPGenc, use multipass vbr. Set your min value to the lowest your player will accept. Make sure your set for interlaced input. Set your output to interlaced as well. You can also try reducing the DC Component precision to 8. This takes up less bitrate, freeing up more for your video. You should also search the forum for "matrix" and "cam". I remember seeing someone post a quantization matrix for Cam sources that was supposed to optimize your output.

[tandr]

vhelp

It is type 2 DV shot -- I grabbed it using Microspoft Movie Maker and AFAIK it hould be type 2

the problem is that it says "Cannot find decomressor for dvsd format ... bla bla" and I dont want to mess with registry settings that affect all machine. Why vdub cant use directshow reader, is beyond my understanding, it should be not that hard for them to implement it...

DJRumpy,

thanks for suggestion of tools. I dont want to make CVD -- they look even crappier on my screen (I hate artifacts on the edges).

lower resolution == it is 2 edged sword. Sounds funny, but the best result that I got so far is when I use full 720x480 resolution for encoding. Iam not looking to put 2 hours in CD-R, 30-40 min will do it for me.

oh, and as side info -- I have Norcent DP-300 DVD player with Dolphin firmware, playes everything tht I throw in it (so far).

[DJRumpy]

The "Cant find decompressor" error indicates a codec problem. You may have to uninstall/reinstall your DV, or audio codec to get around it.

You can also try GSpot to see if the codec is installed and working properly. Find it in the TOOLS section.

As for resolution, remember that the higher the resolution, the higher the necessary bitrate to keep it looking good. I suggest CVD only if you are trying to create a compliant SVCD or CVD. If that's not an issue, than definately stick with the higher resolution. I mentioned it here because you asked specifically about the SVCD samples. CVD is a variant of SVCD. Hope that explains it...

[tandr]

Hoopps.... now you got me Well, I am newb, so it is not that hard

I thought VCD == CVD in your post, but looks like I am wrong big time.

thanks for GSpot, gonna take a look what is that, may be it will help to resolve my vdub story.

just noticed that VDub is GPL, hehe. I am going to take a look in the code, may be can do something there (never worked with DirectX, but what the hell...)

t.

[DJRumpy]

CVD is MPEG-2 (352x480)
VCD is MPEG-1 (352x240)

The taller vertical makes a big difference in output quality. There is no comparison between the two. CVD done properly will look like SVCD, albeit a hair softer. It's also resolution compliant with DVD. Just use the same bitrate settings for it, as you would for SVCD.

[tandr]

just noticed that VDub is GPL, hehe. I am going to take a look in the code, may be can do something there (never worked with DirectX, but what the hell...)

Nah, not gonna do that, found a way to read them in (panasonic's vfw filters). Thanks everyone for help.

I guess I need some filter for that static (constant?) noise reduction.

is it possible to write MPEG2 files from vdub somehow? (offtopic, srry)

[DJRumpy]

No. AVI, Audio, or frameserved output only

[vhelp]

@ DJRumpy..

That was a good explanaa on the bitrate/mpeg encoding analigy

It also helps to make one better understand the whys/dos/don'ts in DV
footage'ing around in your hand. Now, you all understand why Hollywood
uses those crains and other various gizmo contraptions hooked up to their
Video CAMS

I saw something similar, but more compact, and is cariable in the hand.
Sort of looks like an "L" shaped or something, w/ a hand attachment.
Anyways.. says that it give your CAM footage smoother and less wobbles
and no vibrations during shooting - - just what we all need, so that the
above MPEG analigy won't apply as much, and our bitrate will be minimal
(for amature footage'ing around)
Didn't get the price though

-vhelp

[DJRumpy]

Thanks vhelp. One point I forgot to make that I neglected to describe, is the result of all of this 'missed' motion. In the above example, if bitrate has to be used to create an image area, then that bitrate could starve the overall image of necessary bitrate. This happens when the amount of bitrate the encoder has available in any particular second, is insufficient to reproduce a group of pictures in that second. As the encoder misses more an more objects in motion (the farther it gets from the I-Frame), you get a sort of domino effect. When the encoder finally gets to the point, where it's applying all available bitrate to the frame to re-produce image area, and it cannot accurately re-produce an image becuase it lacks the necessary bitrate, it lowers the overall quality for each macroblock area. This results in a 'blocky' appearance (or as most like to call it: macroblocks ).

[vhelp]

@ DJRumpy..

Again, thanks..

Here's something that's ben nagging me a bit.. I'm a litle confused about
how people go on about BLOCKS and BOARDERS (aka, letterbox) and how
one should so call, "align" them at 16 pixel (or is it 16x16 blocks) for the
maximum "even" bitrate distributes. Ok, so in looking at the pic below,
when you see the:

* Picture, Slice, Macrobox and then finally, the block..



..there is a 16x16 block which has 4 blocks in them. Each 4 blocks are 8x8 pixels.

So, in rethinking the maximum pixel distrbutes, is it STILL better to lay
out a 16x16 boundary (when letterboxing around the edges in captures that have
partial blacks and distortions from VCR ect) OR, should I do an 8x8 instead ??
OR, 16x16 or 8x8 is ok eather way ??

I use 16x16 around all my edites (top/bot/left/rt) in my encodes, just to give
my final encoding that much more edge, since in my captures and even DV (advc)
I get some blacks in any one edges or distortions arounds some. So, in my
standard encoding, i just stick to 16x16.

Now, after all that..
In eather event, I'm trying to come up w/ a forumla that will calculate the
black (aka, boarders/letterboxing) from top/bottom to evenly produce 16x16
(or, 8x8) that will fill evenly, between top of black, to bottom of black,
just at the point of where the video portion begins. If that means doing
a slight bit of stretching, then maybe that could be factored and calculated
as well. And, doing a little stretch or shrinking or cutting wont be too serious
to the final video, in order to fit the calculation of laying out the 16x16 blacks.

What do you think about this analigy (or theory) idea of mine ?? ..crazy ??
yeah, but still worth it maybe

If you understand what I'm going on about, maybe we could help eather other
out on this (and ohers can benefit) I vagly recall mentioning this at kvcd.net
some time ago, but I don't think they understood where I was going w/ this.
It's ben while ago since. Anyways.. I do hope that you understand and can
offer up some answers and what-nots.

Thanks a bunch!!
-vhelp

[DJRumpy]

As a rule, your frames vertical, and horizontal should always be divisible by 16. This can encompass macroblock pixels values of 4, 8, or 16. This way, your covered no matter what.

In eather event, I'm trying to come up w/ a forumla that will calculate the
black (aka, boarders/letterboxing) from top/bottom to evenly produce 16x16
(or, 8x8) that will fill evenly, between top of black, to bottom of black

Someone has already done this. DV Tool has a calculator, with an option to increment the width/height by 16.

As for how important it is? Not very. You full frame resolution will always be divisible by 16 (720,704,576,480,352,288,240) assuming your using a standard resolution, so your image will always have the proper number of macroblocks no matter how you slice dice, or letterbox it. If you happen to letterbox in the middle of a macroblock, it's not a big deal. Whatever portion of the macroblock that has letterboxing will simply never see any motion, while the other half of the macroblock might. The encoder will handle this without issue, just like it does any other motion. It only sees the small picture (macroblock). It doesn't care a bit if the macroblock is half empty (black), or half full. It's all just motion...