Auto Gordian Knot: current version 2.55

[Sharktooth]

Installed 1.02... tonigth i'll finish to encode.
V3ULR compressibility is almost the same as Jawor's 1CD.

[rantingotaku]

well, whatever else you changed fixed my problem Len0x (I'm guessing resizer).

I guess a big "thank you" is in order, even though it's more of an accidental fix thanks.

[len0x]

Originally posted by rantingotaku
even though it's more of an accidental fix

This was no accident - I was not satisfied with Lanczos4 myself.
New matrix is clearly better (and will be even more) and removegrain is a very good filter as well. The only variable that was never tested widely is resizer.

[len0x]

Got report that RemoveGrainSSE2 crashes on it. Anyone has this issue?

[Taurus]

Originally posted by len0x
I was not satisfied with Lanczos4 myself.
The only variable that was never tested widely is resizer.

I think this is a good decision. Felt not comfortable with the new resizer, even I am a sharp looking guy

[CeeJay.dk]

I just remembered since you mentioned a CPU specific part of AutoGK ..
http://allies.freezope.org/ host optimized builds of Xvid 1.02 for Athlon XP and Athlon Thunderbird processors that are faster than the ordinary builds.

I haven't done any proper comparisons but I get really good speeds with this optimized build.

[therealjoeblow]

I tried to reencode the movie I commented about a few replies ago, and put .sharpmatrix into the VIDEO_TS directory with the .vobs, but Autogk's log reads:

[10/27/2004 8:45:11 AM] Compressibility percentage is: 44.28
[10/27/2004 8:45:11 AM] Using softer matrix.
[10/27/2004 8:45:11 AM] Chosen resolution is: 640x368 ( AR: 1.74 )
[10/27/2004 8:45:11 AM] Predicted comptest value is: 52.95

It's hard to tell what was actually used because I deleted the other encodes and can't compare. Do you think this just a logging error, or did it for some reason not use the sharp matrix even though I had the control file in place?

[len0x]

Originally posted by therealjoeblow
I tried to reencode the movie I commented about a few replies ago, and put .sharpmatrix into the VIDEO_TS directory with the .vobs, but Autogk's log reads:

How do you put there - manually or with your tool ?
I'd check if the file was really there before encoding started.

[len0x]

Originally posted by CeeJay.dk
I just remembered since you mentioned a CPU specific part of AutoGK ..
http://allies.freezope.org/ host optimized builds of Xvid 1.02 for Athlon XP and Athlon Thunderbird processors that are faster than the ordinary builds.

Yeah, good builds. Too bad I can't detect processors in the installer of AutoGK - otherwise I could distrubute optimized version automatically.

[Sharktooth]

You can't but the user can...

Xvid codec:
|_ Install Xvid x.x (standard)
|_ Install Xvid x.x for AMD Athlon
|_ Install Xvid x.x for AMD Athlon XP with SSE

[len0x]

1) its too much to bundle
2) it might not be as reliable as koepi's (in terms of when they are compiled and continue to be compiled)

[Sharktooth]

Agreed. Also the speed gain is lower than 3%...

[CeeJay.dk]

HKLM\\Hardware\\Description\\System\\CentralProcessor\\0 contain useful keys that will tell what type of CPU processor 0 is ( there are other entries under 1 , 2 , 3 .. depending on how many cpu's are in the system , but they should all be the same )

The optimized codec I use compresses to 206 kb .. probably would be less if it was archived with other builds of Xvid since the files look alike and can thus be compressed to a smaller overall.

But download links to other good builds from AutoGK's download page would also go a long way.

[BigDid]

Originally posted by CeeJay.dk

http://allies.freezope.org/ host optimized builds of Xvid 1.02

I haven't done any proper comparisons but I get really good speeds with this optimized build.

Tks for the link, I tried the xp build.

Done encode on 1 source with hard compressibility to compare. Results are 8% gain for the Xp build on comptest (9.53 to 10.33fps) but marginal on first (33.55 to 33.89) and second pass (24.45 to 24.83).

Hi to all

[len0x]

Originally posted by Sharktooth
Installed 1.02... tonigth i'll finish to encode.

Any news? I tried another sample (Star Wars 2) and again got compressibility that is between V2 and V3HR. This time it was considerably more than V3HR since number of key-framey was 1.7% vs 3.5% that was in my previous test source.

Originally posted by Sharktooth
V3ULR compressibility is almost the same as Jawor's 1CD.

I found it actually greater than Jawor's. Howerver V3UL seems to resolve more details with no particular difference in blocking (I really need to go back back to my first sample where I can see clear blocking even on high bitrate matrixes @quant=2 - its tough to judge otherwise). having said that more details - doesn't mean better picture on low bitrate matrixes. Removing details and making picture smoother (Jawors) might be better approach since details don't look too good comparing to HR matrixes (but V3UL does try to do that though).

[BigDid]

Hello Taurus & Lenox

New to the forum, autogk and encoding I was wondering if the lanczos4/lanczos question applies to my problem ?

Originally posted by Taurus
Is it just me, or do others have the same visual aspect on lanczos4resize?
It seems to me, as if the edges and lines get something overdrawn.
...
But on PC Monitor sometimes it looks a little distorted.

The source is DV cam authored to Pal/dvd, quality not so good. I tried to encode to xvid (messed the sound btw) in AGK 1.70, needed 1/4 dvd for 4.2 Gb.
The result is medium to better but I have distorded lines (jaggies?) in the original and the encode (see pics). I believe it comes from the interlaced material (see log) and tried to change the kerneldeint treshold to 3 or 30 with no improvement.


I have no other clue on the problem and hope you have good suggestions to offer.
Enclosed pictures from the original (mpg2 720w) and the avi (xvid 512w)

Also the log of one of my trials encode:
[24/10/2004 17:27:47] AutoGK 1.70b
[24/10/2004 17:27:47] OS: WinXP (5.1.2600).2
[24/10/2004 17:27:47] Job started.
[24/10/2004 17:27:47] Input dir: F:\FILMS\DVDVOLUME\VIDEO_TS
[24/10/2004 17:27:47] Output file: F:\FILMS\DVDVOLUME\Mariage Heydy.avi
[24/10/2004 17:27:47] Audio: Unknown language
[24/10/2004 17:27:47] Subtitles: none
[24/10/2004 17:27:47] Codec: XviD
[24/10/2004 17:27:47] Target size: 1493Mb
[24/10/2004 17:27:47] Custom resolution settings: minimum width of 560 pixels
[24/10/2004 17:27:47] Custom audio settings: VBR MP3 with average bitrate: 128Kbps
[24/10/2004 17:27:47] Started encoding.
[24/10/2004 17:27:47] Demuxing and indexing.
[24/10/2004 17:29:57] Processing file: F:\FILMS\DVDVOLUME\VIDEO_TS\VTS_01_PGC_01_1.VOB
[24/10/2004 17:29:57] Processing file: F:\FILMS\DVDVOLUME\VIDEO_TS\VTS_01_PGC_01_2.VOB
[24/10/2004 17:29:57] Processing file: F:\FILMS\DVDVOLUME\VIDEO_TS\VTS_01_PGC_01_3.VOB
[24/10/2004 17:29:57] Processing file: F:\FILMS\DVDVOLUME\VIDEO_TS\VTS_01_PGC_01_4.VOB
[24/10/2004 17:29:57] Processing file: F:\FILMS\DVDVOLUME\VIDEO_TS\VTS_01_PGC_01_5.VOB
[24/10/2004 17:29:57] Source aspect ratio: 4:3
[24/10/2004 17:29:57] Source resolution: 720x576
[24/10/2004 17:29:57] Found PAL source.
[24/10/2004 17:29:57] Analyzing source.
[24/10/2004 17:37:55] Source has percentage of interlacing in motion areas: 72,82
[24/10/2004 17:37:55] Source is considered to be interlaced.
[24/10/2004 17:37:55] Found 145848 frames
[24/10/2004 17:37:55] Encoding audio.
[24/10/2004 17:38:11] Audio size: 1,855,488 bytes (1.77 Mb)
[24/10/2004 17:38:11] Overhead: 5,833,984 bytes (5.56 Mb)
[24/10/2004 17:38:11] Video size: 1,557,834,496 bytes (1485.67 Mb)
[24/10/2004 17:38:11] Running compressibility test.
[24/10/2004 17:38:11] Writing the following script to F:\FILMS\DVDVOLUME\agk_tmp\Mariage Heydy_comptest.avs
===========================================================
LoadPlugin("C:\PROGRA~1\AutoGK\DGMPGDec\DGDecode.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\autocrop.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\RemoveGrain.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\kerneldeint.dll")

movie = mpeg2source("F:\FILMS\DVDVOLUME\agk_tmp\Mariage Heydy.d2v")
cropclip = autocrop(movie,mode=0,wmultof=4,hmultof=4,samples=10,aspect=0,threshold=34,samplestartframe=0)
fixed_aspect = 1.06666666666667
c_width = width(cropclip)
c_height = round(height(cropclip) / fixed_aspect)
input_par = float(c_width)/float(c_height)
input_par = input_par > 1.4 ? input_par : (4.0/3.0)
out_width = 704
out_height = round(float(out_width) / input_par)
hmod = out_height - (floor(out_height / 16 ) * 16)
out_height = (hmod > 4) ? (out_height + (16 - hmod)) : (out_height - hmod)
new_aspect = (float(out_width) / float(out_height)) / fixed_aspect
KernelDeInt(movie,order=0,sharp=true)
autocrop(mode=0,wmultof=4,hmultof=4,samples=10,aspect=new_aspect,threshold=34,samplestartframe=0)
Lanczos4Resize(out_width,out_height)
RemoveGrain(mode=2)
SelectRangeEvery(300,15)
===========================================================
[24/10/2004 17:58:12] Duration was: 20 minutes 0 seconds
[24/10/2004 17:58:12] Speed was: 6,07 fps.
[24/10/2004 17:58:12] Compressibility percentage is: 45,15
[24/10/2004 17:58:12] Chosen resolution is: 576x432 ( AR: 1,33 )
[24/10/2004 17:58:12] Predicted comptest value is: 67,45
[24/10/2004 17:58:12] Running first pass.
[24/10/2004 17:58:12] Writing the following script to F:\FILMS\DVDVOLUME\agk_tmp\Mariage Heydy_movie.avs
===========================================================
LoadPlugin("C:\PROGRA~1\AutoGK\DGMPGDec\DGDecode.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\autocrop.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\RemoveGrain.dll")
LoadPlugin("C:\PROGRA~1\AutoGK\filters\kerneldeint.dll")

movie = mpeg2source("F:\FILMS\DVDVOLUME\agk_tmp\Mariage Heydy.d2v")
cropclip = autocrop(movie,mode=0,wmultof=4,hmultof=4,samples=10,aspect=0,threshold=34,samplestartframe=0)
fixed_aspect = 1.06666666666667
c_width = width(cropclip)
c_height = round(height(cropclip) / fixed_aspect)
input_par = float(c_width)/float(c_height)
input_par = input_par > 1.4 ? input_par : (4.0/3.0)
out_width = 576
out_height = round(float(out_width) / input_par)
hmod = out_height - (floor(out_height / 16 ) * 16)
out_height = (hmod > 4) ? (out_height + (16 - hmod)) : (out_height - hmod)
new_aspect = (float(out_width) / float(out_height)) / fixed_aspect
KernelDeInt(movie,order=0,sharp=true)
autocrop(mode=0,wmultof=4,hmultof=4,samples=10,aspect=new_aspect,threshold=34,samplestartframe=0)
Lanczos4Resize(out_width,out_height)
RemoveGrain(mode=2)
===========================================================
[24/10/2004 19:37:17] Duration was: 1 hour, 39 minutes 5 seconds
[24/10/2004 19:37:17] Speed was: 24,53 fps.
[24/10/2004 19:37:18] Expected quality of first pass size: 61,90
[24/10/2004 19:37:18] Running second pass.
[24/10/2004 21:51:27] Duration was: 2 hours 14 minutes 9 seconds
[24/10/2004 21:51:27] Speed was: 18,12 fps.
[24/10/2004 21:51:27] Job finished.

I am very fond of autogk and do appreciate all the efforts improving it. All the best

Did

Oops... the original mpg2 pic is missing, new attachment:

Edit2: no luck with the mpg2 picture

[len0x]

Originally posted by BigDid
The source is DV cam authored to Pal/dvd, quality not so good.

If its already in the source then nothing can be done...
(and for DV sources its quite common to have such ladder step lines if I'm not mistaken).

[Sharktooth]

Originally posted by len0x
Any news? I tried another sample (Star Wars 2) and again got compressibility that is between V2 and V3HR. This time it was considerably more than V3HR since number of key-framey was 1.7% vs 3.5% that was in my previous test source.

Well, in my AGK-HR matrix the more the keyframes the lower will be the compressibility coz the Intra-Frame matrix has lower coefficients than the Inter-Frame matrix. The "key-frame pumping" effect of this kind of matrices ensures a better overall quality (coz I-Frames are reference frames for P-Frames... highering I-Frame quality = boosting overall quality...) but there is a drawback. In a movie that requires a lot of I-Frames the compressibility hit is noticeable.
So i will try to find a good compromise between Intra and Inter-Frame matrices coefficients but it requires some time...
Saturday or sunday it will be ready.
However here's a second try (beta 2):

  8   9  10  11  12  14  16  18
  9  10  11  12  14  16  18  20
 10  11  13  15  17  19  21  22
 11  12  15  17  20  22  24  26
 12  14  17  20  23  25  28  31
 14  16  19  22  25  29  34  40
 16  18  21  24  28  34  44  52
 18  20  22  26  31  40  52  74

 16  16  16  16  17  18  21  24
 16  16  16  16  17  19  22  25
 16  16  17  18  20  22  25  29
 16  16  18  21  24  27  31  36
 17  17  20  24  30  35  41  47
 18  19  22  27  35  44  54  65
 21  22  25  31  41  54  71  86
 24  25  29  36  47  65  86 115

Direct download link: http://ebola.gamersrevolt.it/len0x/eqm_AGKHR_b2.xcm

Originally posted by len0x
I found it actually greater than Jawor's. Howerver V3UL seems to resolve more details with no particular difference in blocking (I really need to go back back to my first sample where I can see clear blocking even on high bitrate matrixes @quant=2 - its tough to judge otherwise). having said that more details - doesn't mean better picture on low bitrate matrixes. Removing details and making picture smoother (Jawors) might be better approach since details don't look too good comparing to HR matrixes (but V3UL does try to do that though).

When you finish to test it inform me about the results.

[BigDid]

Originally posted by len0x
If its already in the source then nothing can be done...

Ok, this is it Thanks for the quick reply anyway.

When I will have the time I will try to dig, learn, use others progs like avisynth filtering, vdubmod with xvid settings but not at the moment, too much at once.

In this context I will appreciate if/when filtering (whatever the method) become available in AutoGK but I can understand that it is not the highest priority at the moment. All my wishes for continuing developping this very good application.

Did

[therealjoeblow]

Originally posted by len0x
How do you put there - manually or with your tool ?
I'd check if the file was really there before encoding started.

I used my tool, but here's time/date/filenames cut and pasted to ensure no errors:

.sharpmatrix
Created: Wednesday, October 27, 2004, 8:19:29 AM

->from the AutoGK.log:
[10/27/2004 8:19:50 AM] AutoGK 1.70b
[10/27/2004 8:19:50 AM] OS: Win2000 (5.0.2195).2
[10/27/2004 8:19:50 AM] Job started.

Filename appears correct, and was created before encoding started.