Problem with Koepi 04102002 Biuld

[Owen]

Koepi,
I have been using your great builds for some months now for one pass real-time encoding at 720-576 PAL YUV2 captured direct from BT TV card.
I have been getting excellent results, but when I installed the 04102002 build I get very poor quality (blocky and with some picture brake up).
I have now installed Nic's current build and all is well again.
My setup is:
One Pass Quant. Mode. Quant = 2
Motion search = High
Interlace = On
Max Keyframe = 25 (for playback flexibility)
All other setting at default.

My system is WinXP Pro, on P4 2Gig

Since no one else seems to have mentioned this so I thought I should.

Hope the feedback is of use and keep up the great work.

Regards,

Owen.

[Koepi]

Try hitting "load defaults" before using a new build the first time.

I'm quite sure that there is _no_ problem with my build but with your setup.

Regards,
Koepi

[iago]

@Owen

I've been using Koepi's 04102002-1 build with great results since the day it was released, though I work only with DVD sources. I've just recently finished two 2-pass full-movie encodes and both are pretty good.

regards,
iago

[Owen]

Koepi,
Thanks for your reply.
I normally reset defaults when installing new builds (Not a newbie)
I have uninstalled Nic’s build.
Installed 04102002-1
Reset defaults.
Captured video.
Played video with and without ffdshow.
Video shows blocking/noise (not normal blocks due to compression)
Uninstalled.
Installed 23092002
Tested as above.
Captured video plays perfectly.
Uninstalled.
Installed Nic’s build.
Tested and it works perfectly as well.
I have never had problems with your old builds.
04102002-1 is the ony one that is a problem.
Please note that I am not complaining about your build, just reporting a possible bug.
Remember that I am CAPTURING with the codec not re-encoding.
Can anyone here capture with my settings to test ?
To all the developers of Xvid, congratulations and my thanks for a great codec.
I look forward to future builds.

Regards,

Owen.

[Koepi]

After installing those other binaries, did you try my build again?

I'll test-capture at fixed quant 2 in some minutes to see if I can verify your result (btw., which resolution do you capture with? my duron700 can't handle full PAL/interlaced in realtime with xvid ;) ).

Regards,
Koepi

[iago]

@Koepi and all,

Anyone experiencing problems (like blockiness, extra noise, and/or quality decline, etc.) in their "two-pass encodes" as well with the 04102002-1 build? I've been (still) using this build without any problems so far, but would like to know such issues if present, in case they might have gone unnoticed by my eyes (?) ;).

If so, maybe we can make Koepi release a new "stable" build! :D

best regards,
iago

[Koepi]

@iago,

2pass works absolutely fine.

This issue here is about 1pass fixed quant :)

Don't scare the hell out of me here ;)

Best regards dostum,
Koepi

[Owen]

Koepi,
I have installed and uninstalled 04102002-1 three times in total and always get the same result.
I have looked at some Keyframes in Vdub and can clearly see the fault.
For example, a horizontal edge between a light area on top and dark area below can show a segment where the light and dark are reversed with dark on top of light below.
Like just a small segment of scan line is swapped with the line below or above.
I can see a similar fault in vertical edges.
It looks very strange.
I can send you some frames to look at if you like.
Give me an address.

Regards,

Owen

P.S. I capture at DVD 720-576 res. Pentium 4 2Gig handles this with almost no dropped frames for hours on end.
I dont compress audio while capturing, to many sync problems.
Pentium 4 2.8+ Gig coming soon, should handle Qpel and maybe sound compression realtime as well, I'll report.

[-h]

Could you post a screenshot to the board? It sounds like interlacing has gone haywire again, but nothing's changed interlacing-wise for over a month now.

-h

[Owen]

Here are the keyframes, as requested.
They are full res. but jpeg'ed.

Regards,

Owen

[yaz]

hi all !

seems not too important but i use 04102002 wout any problem. even hqmod seems to be implemented :-))) i luv it, really, the best beta ever. but ... at some realeses i noticed funny things.
- when i popped up xvid in vdub it gave different defaults than with the command line config. sorry, i can't remember which build(s) was/were that cus i always change to the newest beta.
- with some builds the registry wasn't cleaned on uninstall. i guess, settings are kept there.

@owen : if u changed 'too big' u can't use 'the same settings' cus these days there're too much flip-flopping with options available. my guess, u hasn't tuned minimal kf-interval with consec i-frame interval, thus xvid takes most/all kf as consec giving quite funny results.

y

[-h]

That's interlacing alright.. I guess someone's changed some code.

Hm I'm not sure what to tell you - interlacing works fine for me in all tests.

It'd be best if you could find a segment (captured in MJPEG or huffy), which when compressed to XviD gives these errors. None of my interlaced material gives me incorrect field swaps like your screenshots.

-h

[-h]

If you're using the stable build, you've encountered what I've dubbed the "stupid programmer interlacing test bug" - the field test was performed incorrectly resulting in crud like your screenshot (fields separated even when a later test says they shouldn't be).

This is now fixed. Dev branch has always worked correctly, I just ported the bias check back to stable in a rather dumb manner.

By the way this isn't anything like the big ugly interlacing bug from a month ago, if you encoded something when this current bug existed, you'll notice straight away when you watch it.

-h

[Owen]

Just to clarify.
This encoding problem happens %100 of the time.
Every capture is corrupted.
I will try capturing in mjpg etc. and re-encode.
I will also try encoding without interlacing.
Im very familiar with the interlace bug in older builds.
Most of my captured video is now only playable with ffdshow with bug fix on.

Regards,

Owen

[-h]

So, which version of Koepi's results in buggy output? The stable or dev version? Stable interlacing should have given buggy output 100% of the time, dev interlacing should have worked 100% of the time.

Your old captured content will only work in ffdshow with workaround bugs on, but anything captured with a recent stable build should be completely bad and not viewable by anything (because of my bug I just fixed).

-h

[Owen]

-h
Should be the stable build.
So what you say makes sence.
Filename was Xvid-04102002-1.exe downed from Koepi’s site when it was first posted.
How is Nics code different?
I thought they use the same CVS code base.
Anyway have we sorted it all out?
Do I need to do any more tests?

Regards,

Owen

[kilg0r3]

Yours is the stable built. The development built surfaces as

Latest development (unstable) binary:
XviD-12102002-1.exe (433kb)