capping for svcd?

[funsoul]

H Folks!

Well...matrox has finally blessed us with drivers for 2k that seem to work (in rgb or yuy2) so finally switched from 98se. Changed the soundcard (damn..hate to give up the old montegoII+) and, lo and behold, the raid started working again! wooohooo! So I'm now ready to start capping larger framesize stuff again (which I've wanted to get better at).

The video quality at 704x480 or 640x480 is yuck-looking at this point *groan* I capture with virtualdub in rgb or yuy2 at the framesizes above. (for some reason I could never get huffyuv and avi-io ver 01.23 to cooperate...result on any huffcap was pure, digi-psychedelia).

Source is vhs (ntsc) using the svhs connection so that would explain some things but viewable vids are the goal. The raw caps look a little pixelated and pretty liney (is that interlacing?), especially where there's motion.

Exporting from vdub to a huffyuv file (I really need to learn avisynth!) at 480x480 for working in premiere, I apply the following filters (in order); smart deinterlace (default settings except that for 'select fields only'), unsharp, then cropping to trim the noise off the bottom (cropping all around to maintain final aspect ratio), then smart resize (480x480, precise bicubic).

Again, the file looks decent except that it comes out with what looks like lines. The filtering seems to be fine except for the smart deinterlacing (there's no preview option for that). They're most visible in areas of the frames with a lot of motion....the video looks excellent in the very still scenes.

Is this an interlacing issue or is it something else? Is there really a way to get it looking pristine (given source mentioned above)?

Any and all guidance, hints, tips, tricks, urls and RTFM's ( ) are greatly appreciated.

Thanks mucho,
-funsoul

My setup:
p4 1.4GHz
512mb pc800 ecc rdram
27g ibm primary
matrox g400-tv (with the newly released drivers)
Hercules FortissimoII s/card
promise fastrack100
135g raid0 (does 52meg/s sustained with Sandra sound like what it should be?)
drives: 3x45g ata100 maxtors
win2k

PS- sorry for the long message but I'm so excited to FINALLY get all the hardware working again! hehe

PSS- anyone know of a nice soundcard with s/pidf in/outs AND digital coax in/outs that'll work under 2k? Also...anyone want to buy a turtlebeach montegoII+ soundcard? *grin*

[funsoul]

hmmm....how do I turn the winkyface into a question mark on the front page of the forum? The frown, grimace, etc would be suitable as well but not the grin

[mikiem]

Hi

I'll start with a really stupid question I'm afraid... The reason for the SVCD size of 480 x 480 I thought was to maintain the interlace to begin with for the TV screen. While I realize it doesn't address your question, if the final video is to be viewed on TV, why not leave the interlace, & if it's for the PC, why go with that aspect ratio since it really doesn't offer anything additional except file size? FWIW, I've also found that it seems easier for most PCs to stretch a smaller picture then to calculate and display a larger one, and usually does so at better quality.

RE: the vdub filters, usually want to start off with the crop first as a rule of thumb -- less to work with & gotta be some faster. HUFFYUV is primarily an editing format to my knowledge, and the fields are quite noticable to begin with -- also, don't know if HUFFYUV is seeing your video as frame based since you're going to 480 -- perhaps it's storing fields still, which you see.

I don't really have a usual mode of operation -- depends on the video -- but I've had some success with half size video in vdub using the horiz half size filter going to a 240 height -- does a really nice job, including most of the deinterlaceing too! Rarely I'll run a delinterlace filter afterwards if it shows it missed some frames, and then I tend to use the built in blur.

Usually don't try to eliminate too much noise, maybe filtering at a comparitively low setting, since I usually compress it down to a bitrate between 1 & 1.4 -- the noise seems to actually help avoid ghosting and pixillation quite often, which to me are far more objectionable then the noise itself.

A very recent example was a movie I ripped (the card was out)... Got a very nice picture, but sized small enough to fit on 2 CDs, the thing looked really awful blown up to a viewable size! Had all sorts of jaggies etc from such a sharp picture. I swtiched to a lower quality, noisier codec for my initial avi files, and these in turn compressed down and enlarge wonderfully.

Quite a few times I'll also try the IVT in Vdub to see if the output looks better then a standard deinterlace. As there are several schemes for the original 3:2 pulldown, and patterns can vary within the movie, quite often I'm not enough of a perfectionist to go much further, though I'll again sometimes use deinterlace to pick up anything IVT might have missed. I have noticed that IVT in Vdub will cut off the last few frames, so I wind up padding the video first, then cutting afterwards.

To output a full size frame to TV, where I want to hold onto the fields (to me generating them on the fly with the Marvel isn't as nice), I usually hold any filtering to a very bare minimum -- maybe a crop and resize -- since just about everything I've tried looks worse then the original.

And finally, another rule of thumb is to use bi-cubic enlarging, and bi-linear shrinking. Since you're doing both, might want to give the bi-linear a try.

Ooooops! forgot the sound card... I liked the break out box on the hercules, which uses the same chip as on the Santa Cruz. Alas, very tight budgets interfered, & I picked up a SB Live 5.1 since it was $30 US off -- so far it seems fine, but then I'm not running via. For a general idea on music (which I suppose we should be doing with video) as opposed to gaming soundcards, check out musiciansfriend.com. There are some nice cards available with more then enough connectivity starting for around $300 US.

mikie

[hreintke]

LS,

I had the same experience as mikie when using the new drivers.
I captured using Yuy2 and used the picvideo (Quality 19 and 20) and huffyuv and got captures which had a lot of lines.

After re-enabling the mjpeg HW features using the T-rex procedure I captured using matrox Mjpeg. In this capture the 'line problem' does not exist.

Is the line problem a driver issue or a codec issue ?

My system consists of :

win2k, SP2, PIII 933, SB live, Marvel 400

Kind regards,

Herman

[AndrewDV]

Hi hreintke,

Are the lines you are getting around objects in motion? Are they more long white lines that only occur in a pan shot?

[hreintke]

Hello,

I know I can get lines when I have fast movig shots. But the lines also appear in alsmost standing shots.
I attached an jpg with parts of both capture to show the difference.

Kind regards,

Herman

[chm]

It looks like you have the field order reversed. Try changing the setting for which field is first.
Sorry, but I don't know which setting you would need for your capture setup.

--Chris

[hreintke]

Hello,

I think the only options for field order are when editing a file and not during capture. Isn't the matrox board combining the fields and just presenting frames to the captue program ?

Kind regards,

Herman

[Dr Mordrid]

funsoul,

AVI_IO 01.23???? Time to write Markus for an update

I'm using AVI_IO 03.19 and it works great. This version has fixes for HuffYUV 2.1.1.

Dr. Mordrid

[mikiem]

Hi all

Thanks doc -- have to check my version of AVI_IO when I get off here too.

Regarding the lines, in my example it isn't or wasn't a field issue since I was going with a 24fps vob file as source, and never went above 240 height. With Herman's, I really can't tell to be honest if it's the same problem but it looks the same.

To be fair to HUFFYUV, I got the same results using the picvideo lossless codec, and this may just be a limitation of current windows software. When adding fields to the mix, it gets more interesting cause you have issues of how the fields get stored internally in the file, if the fields were skewed already from 3:2 pulldown etc... And I'm convinced that on more then one occassion deinterlacing has helped simply by blurring the picture.

Encoding a file with a smaller frame size & having the player blow it up to viewing size, what I think is happening is similar to enlarging any picture without doing something like a bi-cubic routine -- you get all sorts of jagged edges which appear as all sorts of nasty lines when the edges of anything in the frame are horizontal. Using a noisier(?) codec has the same effect as anti-aliasing in a graphics editor -- it blurs or blends the edges.

Beside acting as anti-aliasing, I think it helps with later encoding. Going with a high rate of compression, the compressing software looks for anything it can cut out of the video. I *think* that when the picture is a little bit noisier, the encoder see's this as motion, so takes away less of the original picture.

Regardless, the end result after compressing many a video with more & less noise present, is the one with a small amount of noise usually looks better, especially zoomed to larger size.

Going with a larger size frame to begin with I've found that fewer PCs could actually play the video, and if the video was zoomed further, it looked noticably worse then zooming the smaller sized frames to the same viewing size, full screen in both cases for example. This probably shouldn't be so, but nothing is perfect, and in this case that kinda proves it.

Go with what works

mikie

[funsoul]

Hi again folks!

Hmmm...seem to be having some kind of issue over here. Was on vacation but have now done more tests. To eliminate the shrinking/enlarging caused by 704x480 (yuy2) captures (and the video noise that needs to be trimmed off the bottom of the capture), it would seem better to capture at something larger than 480 but I couldn't get yuy2 to cooperate there. It seems like the best would be to capture at around 640x640. Any ideas?

The other thing that's irking me is the whole re-sizing issue. As an example, let's say the capture is 640x480. On the raw capture, there are approximately 20 lines that need to be trimmed off the bottom. To keep things in the proper aspect, I crop 20 off the top, too (so it's now down to 640x440). How can I get this to 480x480 while maintaining the aspect ratio? One thing that I thought could work is to basically make the vid display 480x480 with 20 lines of black border at the top and bottom of the image (like letterboxing). I thought it was vdub's null transform that would do this but it doesn't seem to work the way I hoped it would. Again...hehe...any ideas?

Also...I noticed that a 320x240 RGB capture viewed at 200% (so 640x480) actually looks better somehow than a yuy2 capture at 704x480. I thought this was weird to say the least. Is there a reason for this or is it just my eyes?

Since the visual comparisons look like RGB would be the way to go, have been playing with that. Unfortunately, the maximum size RGB capture my machine can make is 320x240. Any larger than that and cpu usage (virtualdub) spikes up to 100% resulting in massive drops. Sustained throughput is up around 50meg/s so shouldn't be an issue. Am using a p4 1.4GHz, 512meg pc800 ecc rdram and asus p4t combination so I would've thought 640x480 RGB captures would be possible. Any clues?

OK....still have a lot more questions than answers but guess that's the way it is for the time being.

-funsoul

PS- Yah Dr Mordrid...guess it's about time for an upgrade to avi-io! The forced locking of vid/aud sounds like what I need. I've been capping some old bootleg concert videos and finding that many of them drift out of synch

[Flying dutchman]

@funsoul:

if your purpose is to make SVCD's, it is enough to set the aspect ratio to 4:3 525 lines NTSC in your Mpeg encoder. The TV set will display it at the correct aspect ratio.

In order to display it with the correct aspect ratio on a PC you must either install a DVD player software or the latest Ligos mpeg decoder (it automatically stretches the mpegs to the correct aspect ratio in Windows media player)