1.0+Trellis fix request.

[Sharktooth]

Since i have not so much time to play with xvid sources, is there anyone willing to compile a reasonably fast win32 build of 1.0 with Trellis fix?

[celtic_druid]

Ok, just did a release-1_0-branch compile which looks like it contains the fixes.

With and without installer.
http://celticdruid.no-ip.com/xvid/

[Sharktooth]

Thanks a lot.

[amango]

These builds does not work. I get an error message with Virtual Dub "xvidcore.dll not found", after starting the encoding process.

[Sharktooth]

Yep. Got the same problem.

[Teegedeck]

Reinstall your previous XviD and try just replacing the old DLLs with the new ones. I think it should work for the time being.

[celtic_druid]

Just tried it, which I must admit I had not before (had a cvs head compile installed) and it worked fine.

Only tested the installed on XP though. Just download the 7z version and use the inf to install the codec and regsvr32 the .ax or as said simply copy over a 1.0 install.

The error would I guess suggest that the vfw file is installed and registered correctly, but that xvidcore.dll is not found in the same directory or at least not where it is supposed to be.

[Sharktooth]

I only tried the 7zip and used the .inf for installing and got the error.
Reinstalled the Koepi's build and it worked. so i tried to replace the 2 DLL files and got the error again.

The Xvid GUI shows up correctly but an error message comes up exactly when VMD starts the encoding.

[celtic_druid]

What does the about box say? Because usually it reads the xvidcore.dll version.

[Sharktooth]

It says blah blah blah May 31 blah blah... and xvidcore.dll not found

[springl]

Some ICL-compiled files require libmmd.dll in \windows\system32

[Sharktooth]

Koepi's build is built with ICL too.

EDIT: libmmd.dll fixed the shiznit.

[springl]

;)

[celtic_druid]

Think I may have it. No matter where you install xvidvfw.dll, xvidcore.dll will only be found if it is in a path directory whether that be system32, your maya bin directory or whatever.

Even if it (xvidcore.dll) is in the same directory as xvidvfw.dll it still doesn't seem to be found unless of course that dir happens to be listed as a path.

[Sharktooth]

it was libmmd.dll...
damn ICL.
Btw thanks for the build:) i hope trellis got really fixed.

[Leak]

Originally posted by celtic_druid
Think I may have it. No matter where you install xvidvfw.dll, xvidcore.dll will only be found if it is in a path directory whether that be system32, your maya bin directory or whatever.

Even if it (xvidcore.dll) is in the same directory as xvidvfw.dll it still doesn't seem to be found unless of course that dir happens to be listed as a path.

Still, without libmmd.dll it's a no-go; thank god I had that file lying around because of HeadAC3he.

Sou you might either want to distribute it in your installer/zip or link libm.lib statically into your binaries if they should work right out-of-the-box...

np: Sami Koivikko - Nite Herbs

[zettai]

Is there any way of compiling this without the need for libmmd.dll?

[chilledoutuk]

celtic_druid has libmmd.dll on his site i dont see the problem. cheers for the builds by the way celtic_druid

[pernas]

Thanks celticdruid :)

Has anyone tried the fix yet ?

/Per

[Soulhunter]

Originally posted by chilledoutuk
celtic_druid has libmmd.dll on his site i dont see the problem. cheers for the builds by the way celtic_druid
RareWares also... (http://www.rarewares.org/files/libmmd7.zip) ;)


Bye

[Sharktooth]

Originally posted by pernas
Thanks celticdruid :)

Has anyone tried the fix yet ?

/Per
Yes, but im still trying to understand the results.
Sometimes it gets higher PSNR (trellis enabled) but other times koepi's build is better.
Weird.
Im also trying to identify when (what situations) it is better and when it is not.

[Teegedeck]

What do PSNR values tell you anyway?

[yaz]

@celtic-druid
which is 'that' compile ? & is the 1.1ver is also corrected ? what is that brand new 1.0.0. compiles being also there ?

@sharktooth
what way do u test psnr ? i'm looking for a 'quick' way to do it. that xvidvfw hack found sw here does not work for me.

thx
y

[Sharktooth]

I'm doing all tests with PSNR Lab 1.1.3.
Test material: Spiderman 2 trailer (DVD source)
Avs script for testing:

LoadPlugin ("C:\SOFTS\AVISYNTH\PLUGINS\MPEG2DEC3DG.DLL")
MPEG2Source ("C:\SP2TR\sp2tr.d2v")

Bitrate: 1500kbps
Test mode: High motion
Results are coming in few hours.

[celtic_druid]

The 1.0.0 EXE is new because I added libmmd.dll to the installer, nothing else changed other than the icon. Install it and you will see the build date is not changed.

Yes the head branch (1.1) contains the trellis fix as well as bframes fix, etc. All fixes will I would imagine be ported over and have been so far to my knowledge.

Not sure what you mean by "that" compile.

cvs.head is what will become 1.1
release-1_0-branch is exactly that, the 1.0 release branch.
The EXE's contain the same files as the latest 7zip files.

[yaz]

@sharktooth
... psnrlab ... got it, thx :-) having completed some short tests i have the same impression about trellis. sometimes it seems better sometimes ... more tests needed. however, that strange 'softening' effect seems to be away.

@celtic_druid
thx for the great compiles. i meant by 'that' what u answered :-))

have u given up providing that great mencoder snapshots ? i loved that

thx guys
y

[celtic_druid]

I could do a mplayer/mencoder compile if you like. They are specific AthonXP compiles. Reason being that they were all basically compiled for personal use. I just compress them and drag them over to my server.

[yaz]

Originally posted by celtic_druid
I could do a mplayer/mencoder compile if you like. They are specific AthonXP compiles. Reason being that they were all basically compiled for personal use. I just compress them and drag them over to my server. khmm ... khmm ... i use your compiles (aac, ffmpeg, mencoder, xvid, ... ) for months & yes, i liked them. hope u don't mind :-) i just asked for menc cus since your 'last release' there were 3 new cvs releases at mplayerhq.

sorry for being ot. i've stopped :-))

the bests
y

[Sharktooth]

Here's the psnr lab results for celticdruid's build.

none: 39.6539
b-frames: 39.9322
trellis: 39.8337
trellis+b-frames: 40.1070
cm: 39.7169
cm+b-frames: 40.0319
cm+trellis: 39.8757
cm+trellis+b-frames: 40.1887
vhq4: 40.3151
vhq4+b-frames: 40.4700
vhq4+trellis: 40.4089
vhq4+trellis+b-frames: 40.5872
vhq4+cm: 40.3185
vhq4+cm+b-frames: 40.4798
vhq4+cm+trellis: 40.4211
vhq4+cm+trellis+b-frames: 40.5856
gmc: 39.6447
gmc+b-frames: 39.9136
gmc+trellis: 39.8208
gmc+trellis+b-frames: 40.0908
gmc+cm: 39.7118
gmc+cm+b-frames: 40.0067
gmc+cm+trellis: 39.8855
gmc+cm+trellis+b-frames: 40.1866
gmc+vhq4: 40.3378
gmc+vhq4+b-frames: 40.4838
gmc+vhq4+trellis: 40.4347
gmc+vhq4+trellis+b-frames: 40.5855
gmc+vhq4+cm: 40.3371
gmc+vhq4+cm+b-frames: 40.4882
gmc+vhq4+cm+trellis: 52.7371 (bug in compareyv12: only 37 frames processed)
gmc+vhq4+cm+trellis+b-frames: 40.5949
qpel: 39.1280
qpel+b-frames: 39.4344
qpel+trellis: 39.2718
qpel+trellis+b-frames: 39.5998
qpel+cm: 39.1959
qpel+cm+b-frames: 39.5583
qpel+cm+trellis: 39.3456
qpel+cm+trellis+b-frames: 39.7039
qpel+vhq4: 40.0099
qpel+vhq4+b-frames: 40.1854
qpel+vhq4+trellis: 40.0863
qpel+vhq4+trellis+b-frames: 40.2676
qpel+vhq4+cm: 40.0112
qpel+vhq4+cm+b-frames: 40.1815
qpel+vhq4+cm+trellis: 40.0985
qpel+vhq4+cm+trellis+b-frames: 40.2818
qpel+gmc: 39.2089
qpel+gmc+b-frames: 39.4808
qpel+gmc+trellis: 39.3590
qpel+gmc+trellis+b-frames: 39.6555
qpel+gmc+cm: 39.2949
qpel+gmc+cm+b-frames: 39.6113
qpel+gmc+cm+trellis: 39.4387
qpel+gmc+cm+trellis+b-frames: 39.7613
qpel+gmc+vhq4: 40.0511
qpel+gmc+vhq4+b-frames: 40.2061
qpel+gmc+vhq4+trellis: 40.1484
qpel+gmc+vhq4+trellis+b-frames: 40.3256
qpel+gmc+vhq4+cm: 40.0680
qpel+gmc+vhq4+cm+b-frames: 40.2206
qpel+gmc+vhq4+cm+trellis: 40.1638
qpel+gmc+vhq4+cm+trellis+b-frames: 40.3174

Trellis always give better PSNR results but look at that "52.7371". PSNR lab bug or what?

EDIT: There was a problem while comparing the original clip and the encoded one: only 37 frames were processed then the script stopped to update the logfile.

I can also confirm the softening effect is now gone.

[Prettz]

PSNR doesn't mean anything. The only thing that matters are the percieved quality of the clip and whether there are any visible artifacts (IMO you should encode an entire clip so that you get the 2-pass bitrate distribution you would normally get, because that's where trellis is supposed to help; I never trust the quality of a short encoded segment).

Also, what quantization matrix are you using for that test?

[Sharktooth]

It is a two pass encode of the spiderman 2 trailer.
all the settings are made by PSNR lab so the quantization matrix should be the default (h.263).
I agree PSNR doesnt mean much, but the PSNR variation does mean something.
In Koepi's build test (not yet ready) using trellis does NOT ALWAYS rise the PSNR.
This test was done to see if trellis got properly fixed in this build.

[superdump]

What's the filesize of that seemingly erroneous one? Is it comparable to the others?

[springl]

@Sharktooth
could you please specify which Xvid file you used of those
in http://celticdruid.no-ip.com/xvid/

Thanks,
springl

[CruNcher]

@Sharktooth

10dB difference is most unlikely hmm can you check the output is it ok ?

[Sharktooth]

Originally posted by superdump
What's the filesize of that seemingly erroneous one? Is it comparable to the others?
Same size.

Originally posted by springl
@Sharktooth
could you please specify which Xvid file you used of those
in http://celticdruid.no-ip.com/xvid/

Thanks,
springl
This one: xvid.release-1_0-branch.2004.06.01.7z

Originally posted by CruNcher
@Sharktooth

10dB difference is most unlikely hmm can you check the output is it ok ?
Output is ok and i can't see soo much difference from the other encodes.
I think it could be a bug in compareyv12.dll or in avisynth.
EDIT: Its a bug. i found that in gmc-vhq4-cm-trellis only 37 frames were processed by compareyv12.

[Emp3r0r]

In the new builds I'm seeing a flicker that I've never noticed before. I've looked at my previous 1.0 encodings and can't spot an issue with them... just since the trellis fix. My old encodings used trellis and they look fine.

I've uploaded a short clip that shows the flicker problem. The whole picture doesn't flicker, just very small blocks are flickering. If you view clip, pay attention to the dashboard as there seems to be little flickering blocks.

I'm using default settings with VHQ4, trellis, adapt quant, HVS Better and 3 bframes.

Xvid is the decoder with no PP turned on.

http://jvance.com/files/flicker2.avi (860KB)

[CruNcher]

i think what you recept as flicker is actualy the quantizer change between 2 p and 4 b p is slightly sharper were the b frames are slightly blurrier and you recept that as flicker this 2 4 2 4 2 4 and if you remember that without the trellis fix it wasn't as sharp this could be the reason you know recept that "detail improvement" this way as flicker :) because a blurry more blury change is not at all as recieveable as a hard sharp blurry change this is only a thesis about your problem. This change even could be recieved i think some cm away of a good TFT monitor were on a older TV it shouldn't be anymore recieveable. Yes i can see "recive" this change on my TFT in 70cm distance. You could try my Extra Detail Public Build :) if it's still their i think i know what this b-frame blurryness is coused by and would make a test build for you to check :)

[Emp3r0r]

I think what you recept as flicker is actualy the quantizer change between 2 p and 4 b p is slightly sharper were the b frames are slightly blurrier and you recept that as flicker this 2 4 2 4 2 4 and if you remember that without the trellis fix it wasn't as sharp this could be the reason you know recept that "detail improvement" this way as flickerI need to do some more tests. I'll try turning off trellis and b-frames and maybe rollback to the pre trellis fix builds and see if I can reproduce the flicker and maybe narrow down the cause to what you say above.

[tedgo]

Which from celtic_druid's builds contains the fixes?
The 1.0.0-EXE or the release-branch or both?
Sorry, but i hadn't understood the difference between these two releases...

Btw. I noticed that flickering too

EDIT: I made some tests now, and the flickering only seems to occur with mpeg-matrix, when trellis and adaptive quantization is enabled. I never used adaptive quantization in formerly builds, so it could probably cause by that function that i never noticed it before.

EDIT AGAIN: My question has been answered. Celtic_druid has now renamed the release-branch to XviD 1.0.1.exe. Now i know the difference ;)