Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion. Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules. |
24th February 2011, 18:05 | #361 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
I assume that post refers to the FFTW3.dll issue above... FFTW3.dll is not itself an avisynth plugin. It's a module used by FFT3DFilter. You can't use LoadPlugin on it. Using SetWorkingDir alone will work (as long as no plugin or other script changes the working directory... )
Last edited by -Vit-; 24th February 2011 at 18:10. |
25th February 2011, 02:35 | #363 | Link |
Registered User
Join Date: Nov 2009
Posts: 2,361
|
@fusebyte: be a bit more thankful by turning the complain into a petition. Its a hard work of one guy for free, but GUI? this is avisynth...
__________________
i7-4790K@Stock::GTX 1070] AviSynth+ filters and mods on GitHub + Discussion thread |
25th February 2011, 03:11 | #364 | Link |
Registered User
Join Date: Mar 2004
Posts: 889
|
I just notice (or re-discover) a pattern that QTGMC() fails but TGMC() works.
Pick a true intelaced footage (say, NTSC) that has a solid black 1-pixel line at row 478 on every frame. After bobbing, QTGMC() will have the black removed on even frames, and a 2-pixel-high black line on odd frames. TGMC() keeps the 1-pixel black line on every frame. |
25th February 2011, 05:57 | #365 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
You didn't note your settings [or did you literally use QTGMC()?].
No time to check right now, but maybe Border=true or SVThin=1.0 will restore same behavior. They are defaults changed from original TGMC [Why? My material never needs Border=true, it can cause occasional artefacts of its own and is a *tiny* bit slower. SVThin - I just don't like the results] Last edited by -Vit-; 25th February 2011 at 06:14. |
25th February 2011, 07:11 | #366 | Link |
Registered User
Join Date: Mar 2004
Posts: 889
|
It's TGMC(tr2=3,EDIMode="NNEDI3") vs QTGMC(Preset="Slower",tr2=3). I also notice similar behaviour in sadie's DV sample. His DV seems to leave the topmost line black in some scenes, and after QTGMC(), the black line will disappear and then duplicate on alternate frames.
|
25th February 2011, 16:21 | #367 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
I looked at sadie's example - I don't see a topmost black line - I see a darker line on the second line. And, yes it dances on QTGMC defaults - and the difference from TGMC is that you need Border=true.
The QTGMC docs say of Border: "set true you see flickering on the very top or bottom line of the output" (yep I guess that needs to read 'lines') There's section at the end of the docs comparing with TGMC which also answers this question. You set edimode="nnedi3" on TGMC, only modded versions of beta2 support nnedi3, in the original version that will just do a bob. So let's be exact: TempGaussMC_beta2() is the same as: TempGaussMC_beta2( 2,2,1, 4,0,4, "eedi2" ) is the same as: QTGMC( preset="Slower", edimode="eedi2", border=true, svthin=1.0, lsad=1600, precise=true ) [Corrected - thanks Taurus] The other default differences: Precise switches on/off barely visible steps for a small speed-up. lsad is different because the MVTools documentation is especially confusing on whether that setting should be scaled by blocksize - I had to read the MVTools source code to find that it shouldn't be. Last edited by -Vit-; 28th February 2011 at 04:25. |
27th February 2011, 19:57 | #369 | Link | |
Registered User
Join Date: Mar 2002
Location: Krautland
Posts: 903
|
Quote:
Are you sure TempGaussMC_beta2() uses "nnedi2" as default? From the TempGaussMC_beta2.avsi: Code:
# Defaults for EDI interpolation EdiMode = default( EdiMode,"EEDI2") |
|
28th February 2011, 04:27 | #370 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
Taurus: You're right! I changed my version a long time ago to default to nnedi2, and forgot. I've edited the post above. The equivalence is still there, QTGMC supports EEDI2 also.
nhope: Thanks for that. The presets gradually change. The upcoming version has some noise-related changes because I've brought in [motion-compensated] dfttest. |
28th February 2011, 09:07 | #371 | Link |
User of free A/V tools
Join Date: Jul 2006
Location: SK
Posts: 826
|
@ nhope & -Vit-
Would be nice if you could upload that preset table and made it available on GoogleDocs (not necessarily editable by others) I'm sure it could be helpful for further development/discussion/changes tracking of this wonderful QTGMC script, thanks! |
28th February 2011, 21:49 | #372 | Link |
Registered User
Join Date: Nov 2004
Location: Poland
Posts: 2,843
|
3.11 works much more stable.
About 30fps for 60i SD (slow preset) on 2x X5450. 8 cores at 90%. Used on 60i footage, which had real 60i, badly deinterlaced 60i, 30p and 23.976+pulldown- hahhaha. Final result 29.97p upscaled to HD plays smooth Amazing piece of script Thanks Vit. Andrew Last edited by kolak; 28th February 2011 at 21:52. |
1st March 2011, 05:00 | #373 | Link | |
Registered User
Join Date: Apr 2009
Posts: 478
|
Quote:
|
|
1st March 2011, 08:49 | #374 | Link |
partially-informed layman
Join Date: Jan 2002
Location: Bangkok, Thailand
Posts: 314
|
@ -Vit- - I'll try to remember to go back and update it when your new versions come out.
@ kypec - Version with "faster" as the benchmark & version with "medium" as the benchmark. I just changed the colours to make it easier to visualise the effect of changing the presets, depending on your preset starting point. |
1st March 2011, 12:43 | #377 | Link | |
Registered User
Join Date: Nov 2004
Location: Poland
Posts: 2,843
|
Quote:
setmtmode(mode=5, threads=8) avisource("R:\SD.avi",audio=false) converttoyv12(interlaced=true) setmtmode(mode=2) QTGMC( Preset="slow", EdiThreads=4) converttoyuy2() selecteven() EdiThreads=4- tried different values- nothing helps. Any idea? X5450 are quite old- way slower than i7. I was using avisynth 2.5.7.5 MT dll. Updated to 2.5.8 and now it's crashing much more Andrew Last edited by kolak; 1st March 2011 at 12:46. |
|
1st March 2011, 12:43 | #378 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
nhope: Firstly you can do the horizontal resize (to 1280) before the QTGMC call for a good speed up without loss, just do the vertical resize afterwards.
Distributor() goes at the end of the script. You may not need it - it depends what encoding GUI you use. Try it without. |
1st March 2011, 13:22 | #379 | Link | ||
partially-informed layman
Join Date: Jan 2002
Location: Bangkok, Thailand
Posts: 314
|
Quote:
Quote:
|
||
1st March 2011, 14:54 | #380 | Link |
Registered User
Join Date: Jul 2010
Posts: 448
|
All avisynths resizers operate in two passes anyway (horizontal and vertical). So you don't need to do anything different if you split it up like this and there's no performance penalty from resizing twice in this particular way.
MeGUI requires Distributor(). The MT docs in the OP and in the script apply best to those with 4+ cores. With less cores there is less scope for MT tweaking, especially at HD, but if you ever see your CPU is not close to 100%, then you should see if you can get more out of MT (unless you have stability issues). kolak: Try all versions of MT avisynth, everyone seems to have different stability results. Some find 2.6MT works well. Last edited by -Vit-; 1st March 2011 at 14:59. |
|
|