GOP tweaking to improve AVI -> SVCD quality? (skate video

[dib]

Equipment:
AMD 1.2ghz Thunderbird / ATI AIW 7500 capture card / Maxtor 7200 80gb HD / FIC AZ11EA mobo - All ViaTech 4-in-1 and ATI capture drivers current.

Here's the scenario:
I am capping an old skate video (Powell Peralta Bones Brigade Video 2 - Future Primitive) with composite input and converting to SVCD -- I've had fair to good results working with Intervideo WinDVD-R (which is convenient, but yields unacceptable when played on my Mintek 1600)

Direct captures with the AIW7500 using WinDVD-R with a custom MPEG2 profile (2700 bitrate @ 480x480 / 192kb mpeg2 audio - with GOP = 2 / SubGOP = 1 - motion vectors at 16 /16 or 32/32 (doesn't seem to matter)) still yield ghosting (blurring) on high motion scenes -- which, of course, there's plenty of when your capturing a skate video...


So since the lazy method (heheh) isn't working, I'm looking to capture and convert *properly* using VirtualDub to TMPGEnc, but I've found the pre-made templates - well, *lame*...
I have a 22mb VDub capture (640x480 using PicVideo codec @ quality 19 - few dropped frames) ready on my HD awaiting conversion...

No problems with frameserving the multipart AVI to TMPGEnc, but I tried CCE with dismal results...


I've tried the standard TMPGEnc MPEG2 template with *poor* results - tons of block noise and motion blurring - I've tried softening block noise @ 35... no better...

Encode time isn't an issue -- I'm more than willing to slave out a machine for 8-10 hours doing a proper (clean) encode.

Question: Am I on the right path?
I found that tweaking the GOP/SubGOP settings in WinDVR helped eliminate 'blocking' on the custom MPEG2 profile I set up, so my thought is that this can be applied to TMPGEnc as well. Higher settings produced more blurring though... which I'm trying to reduce (or eliminate)...

Problem is, GOP settings are a bit over my head at the moment - point me in the right direction anyone?

Obviously, I'd like an SVCD-compliant MPEG2 conversion @ 480x480 where the high motion scenes are crisp -- DVD quality is not expected, but I'd like it to be watchable -- How is this possible?

Thanks for reading by the way...

[DJRumpy]

First things first. Clean your video, before encoding. If it's captured through composite, then it's likely to have a lot of noise. Use a good temporal smoother. This noise will eat up your bitrate, making any motion scenes likely to produce macroblocking. You didn't mention how your recent capture looks using PicVideo. If you have the space (i'm guessing you don't, since you captured in segments, or no NTFS support?), you should use Huffman for the highest possible quality capture.

If your source is VHS (you didn't specify), then you should also consider clipping the edges of your video. VHS often leaves a noisy edge, which will eat up bitrate.

If your using CCE, make sure you select the 'Very Low Bitrate' matrix in the Quantize Matrices setting. Use Multipass VBR to encode your video, with a minimum of 0, and a maximum of 2524 (higher if you encode your audio at less than 224Kb/s. Your Average setting will, of course, depend on how big your output can be. If your video is longer than 40-45 minutes, consider using multiple SVCD's to hold your output.

[dib]

Thanks for responding DJRumpy,

Yep, still running Win98se for the moment as this is just a 'slave' system for conversions -- I plan to set up another system dedicated with the cap card and install Win2000 or XP (suggestions?) so I can capture in NTFS.

More information on the VHS cap: Length is 53 minutes total... Source tape is an undamaged, new-old-stock commercial tape... Caps with either Huffy or PICVideo codecs at maximum quality are good with little frame loss - I stuck with PICVideo so I could do some 'smaller' test AVIs and figured I would be going back to Huffy once I got the conversion process a bit more 'streamlined'.

For the moment the segmented AVI is 'soft' when capping from the composite input and I planned to pick up a Sima SCC box to sharpen the captures before input to the ATI 7500... Perhaps even invest in another VCR with S-Video outs for better clarity, budget permitting.


Edge clipping experiments on TMPGEnc last night to help pick up some processing time... thanks!

I had tried CCE recently, but the interface is a little confusing (like TMPGEnc isn't?) -- and it would crap out on me at 12% every time -- after creating a couple large temp files on the HD (I didn't scribble down the file names, sorry).

So I went back to TMPGEnc, knowing that I'm missing out on the superior quality and speed of CCE... Presumably, I need to learn AVISynth to 'join' the segmented files, since frameserving to CCE hasn't been very successful. (Is that how it works?)

Thanks very much for your input - I appreciate it, btw.

[DJRumpy]

On a 55 minute clip, I'd would have to try multipass VBR mode, with a lower AVG setting (use a bitrate calculator...the one from this site is good. Check the tools section). If the quality is still poor, drop back to two discs.

The AVISynth script is pretty simple. It would look something like the example below. Although there are a multitude of ways to write it:

AVISource("clip1.avi","clip2.avi","clip3.avi") #you get the idea...

Save your text in Notepad, and give it a .AVS extension. Double click the .AVS file, an select Windows Media player, or Notepad as the default program. It should play as if it was the original source.

I've never used AVISynth to combine clips this way, so I don't know that the above command works (happy NTFS user). Someone else should be able to help you if it doesn't.

Another easier to read method is to write it so it looks like this:

clip1=AVISource("clip1.avi")
clip2=AVISource("clip2.avi")
clip3=AVISource("clip3.avi")
clip1++clip2++clip3

For the moment the segmented AVI is 'soft' when capping from the composite input

Make sure this isn't a side affect of the PicVideo codec. Try a small huffy clip (very small in your case). You can also try something like MJPeg if PicVideo isn't delivering good quality.

If your video still appears soft, you can also sharpen it up, using avisynth and the Sharpen command. It looks like this:

Sharpen(.03)

The .03 value is a good starting place. A value of 1.0 is the maximum if I remember correctly, although you can string multiple sharpen commands together like so:

Sharpen(.01)
Sharpen(.01)
Sharpen(.01)

Try a few different ways out to see what output looks good. I rarely ever go above .03-.05

Last but not least, the crapping problem in CCE. Does this happen on any type of encode (CBR, VBR, 1-Pass VBR), or just on the VBR modes? I've seen a lot of 'patched' versions of CCE, which crash on VBR mode. Try a quick run in CBR mode to see if it crashes.

If the interface confuses you, don't feel bad. CCE is not radically different from TMPGenc. I happen to think it's a bit more logically layed out. Your mileage may vary....

The settings look different, but perform the same functions as TMPGenc (albeit much faster).

[dib]

Okay - so here's an update --
I've abandoned TMPGEnc for the moment and am concentrating on CCE (now running 2.66.01.07 -- the previous patched version uninstalled )

Now I have a new problem: No audio on the finished mpg.

I've been able to convert from the multisegment AVI capture redone with Huffy Codec at default settings @ 4-pass VBR 1850 avg / 300 min. / 2530 max -- and it looks GREAT!
I can play the .mpa separately and have tried multiplex of .mpv/.mpa in TMPG - still no audio in the finished .mpg

I am not able to select any other compression settings when changing the audio settings in Virtualdub -- just shows PCM Uncompressed.

This is incidentally an issue when capturing too... I have no 'compression' settings - so I just capture PCM uncompressed 256k / 44hz...
Wasn't previously a problem since I've been capping with either PicVideo or Huffy codecs, but I figured it might become one.

I've tried uninstalling / reinstalling all capture and encoding codecs - currently have the 'full' NimoCodec pack installed... Updated my (onboard) AC97 drivers off viaarena.com... Tried uninstalling / reinstalling all video and audio devices in the DM and updated ATI's driver sets...

I've even gone as far as extracting a .wav file from the AVI and setting that as the source in CCE, but it errors at the end of the first pass --
pass #1 - 96%
cce encoding failed: CBR VBV overflow
frame# 89318 (01:49:37:10) | 16856(2.2) max 15570.38(2.0) rel 1285.62 1.08 qsv 112.00->112.00

Starting to feel the pressure -- heheh


Any advise / fixes? (Reinstall the OS?) -- Audio compression settings had worked in the past, but I have no idea when this stopped working (too many experiments / programs later) --

Thanks to anyone brave enough to field this one...

[DJRumpy]

I've had simular wierdness happen before. Trying playing the multiplexed video in your DVD Player software, and see if it plays the video. I had one video I converted, that no matter what I did, the audio would keep dropping out. If I played it in WinDVD, it played fine. It also burned fine.

The culprit turned out to be a codec conflict.

[dib]

Well, here's the latest update -- Biting the bullet and trying to learn AVISynth input using VirtualDubMod -- Now working on a cropped, unfiltered, resized to 480x480 AVISynth script (no problem linking the segmented files -- thanks to DJRumpy!) -- it's far easier than it first appears to be - reminds me of the old BASIC days with the Vic20 (oops - dated myself a little there)...

Will followup later with results of 2/4 multipass VBR encodes using CCE --
All seems well, just *very* time consuming process... Tests have been positive image-wise - Clear as the original AVIs with a little blurring / blockiness - liveable.

Thanks again for your interest DJRumpy - I truly appreciate it.

[DJRumpy]

No problem. Let us know how it turns out. As for AVISynth, it just looks a little scary to the uninitiated. Once you actaully get into it, it's all rather logical in layout and form. Easy as 1, 2, 3...