QTGMC Deinterlacing Script (v3.32) - Page 74

thebigmunch · 16th March 2013, 14:02

Quote:

Originally Posted by LigH

I read about bug fixes inside the code.

So I would vote for "yes". And I know I am not the only one.

The first post is not updated because the author of this update is not the thread starter.

Not so sure about that

Stormborec · 16th March 2013, 14:32

Just two questions:

1)
Is the sharpening function in qtgmc script comparable with lsfmod? At first - there is no supersample in qtgmc - is it a good idea to disable sharpening -> smode=0 ; then upscale and then to use lsfmod?

2)
Is it a good idea to upscale first, if using inputtype=1? It seems, it's preserving more fine details ...

LigH · 16th March 2013, 19:23

Quote:

Originally Posted by thebigmunch

Not so sure about that

Then let's poke -Vit- about it...

osgZach · 17th March 2013, 02:14

I'm not really tweaking anything but speed presets, and I really like the results I am getting on a particular source. The problem is I am getting line artifacts from previous frames showing up very faintly in some parts. It happens in both static and motion scenes, and I'm not sure what I can tweak to try and get rid of it..

It's not always entire outlines of entire objects per-say, just some faint remnants that show up on certain parts of the new frame, in the place where they were in the old frame.?
I guess some kind of blending or residual image. I believe it is also the case that immediately following these frames with artifacts, a much cleaner frame which is almost always an exact duplicate seems to appear in the stream as well.

edit: I think the source may have field blends, no idea if its a conversion though. But I'm a little rusty on procedures used to identify field blending, however. But it does appear artifacts showing up are coming from both C + P frame and possibly N frames too. There do seem to be progessive frames sandwhiched between most if not all of them as well. You can usually tell something is up because the frame image overall does not but there are subtle shifts resulting in some lines looking thinner/thicker and stuff like that..

No idea what to make of this...will post a sample later when I have time, I guess. But any ideas to try in the mean time would be nice.

kypec · 21st March 2013, 09:32

Dear fellows,

I've been using QTGMC with great results so far but recently I spotted very annoying artifacts during normal playback of my encoded footage (PAL 25fps).
The source sample (AVS + D2V + M2V) is available for download here (43MB). Download also fixed AVS + D2V and replace original files that point to non-existent VOB file.

Four consecutive frames that demonstrate my problem below - please note how every other frame is blended, this is actually causing very ugly stutter or judder(?) during normal watching of encoded deinterlaced progressive video @ 25fps.

I've been trying to fight this with few ShutterBlur, ShutterAngleSrc, ShutterAngleOut parameter combinations but achieved no visible improvement apart from slightly blurred frames that were originally non-blended. The blended ones remain intact

Can anyone recommend any possible solution to this? Whatever advanced plugin/script that could help? I don't mind if it's slow as long as it doesn't damage otherwise clean parts of deinterlaced video where QTGMC provides excellent output (watch the rest of the sample opening scene). Only these panning scenes are irritating me. Perhaps I'll try to encode source as x264 interlaced and see how my TV deinterlaces this on playback if no viable solution via Avisynth can be found...

P.S. I prefer to stick with 25fps output (single-rate) if possible because I'm afraid that double-rate (50fps) would be too big for final x264 encode.

osgZach · 21st March 2013, 09:51

fyi, your d2v was pointing to the wrong file.. I just reindexed it. I am guessing the original source was 23.976 NTSC and it went through a PAL conversion, so maybe you need to consult QTGMC's documentation for how to deal with NTSC-> PAL conversions. I don't know much about dealing with them myself. But there are also tools like Srestore to look in to as well.

Didée · 21st March 2013, 09:54

The sample in the ZIP is a fieldblended PAL-to-NTSC conversion. To get progressive 25fps again, use Srestore after bobbing to 60fps .

kypec · 21st March 2013, 10:37

Quote:

Originally Posted by Didée

The sample in the ZIP is a fieldblended PAL-to-NTSC conversion. To get progressive 25fps again, use Srestore after bobbing to 60fps .

Didn't you mean to say NTSC-to-PAL conversion? Sure, this TV series was shot & produced in US originally (NTSC land) and then transferred to DVD Region 2 (PAL land) which is the source I have to work with. Do you mind to show me a script how to bob it to 60fps as the source is 25i, pretty please?

kypec · 21st March 2013, 10:39

Quote:

Originally Posted by osgZach

fyi, your d2v was pointing to the wrong file.. I just reindexed it. I am guessing the original source was 23.976 NTSC and it went through a PAL conversion, so maybe you need to consult QTGMC's documentation for how to deal with NTSC-> PAL conversions. I don't know much about dealing with them myself. But there are also tools like Srestore to look in to as well.

Sorry for messed up index files to everyone. Here are fixed versions of D2V & AVS zipped together, please download and replace.

Didée · 21st March 2013, 11:23

Oops sorry, it is indeed 25i. I just looked at the mediplayer's quickinfo field, and it falsely reported the source framerate being 29.97.
Important point is that QTGMC is not at fault, because the source already contains all those blendings that you issued.
Anyhoo, the strategy stays the same. Just bob it, und stuff it into Srestore.

--EDIT--
However, it seems that the source requires to be restored to 25fps directly, instead of Srestore's default of 23976 for PAL input.

source
bob
Srestore(frate=25.0)

In fact, the scheme of blending is quite similar to the one that is typical for PAL-to-NTSC conversions. (Oversimplified: there are not quite as much blended fields, but a lot more quasi-duplicated fields.)
This is also why I "trusted" the false report of the source being 29.97 fps: because the blending pattern looks like in blended NTSC.

bcn_246 · 21st March 2013, 21:08

Can anybody mod srestore to use SVP?

tia

Didée · 21st March 2013, 21:14

Why, how, what? Srestore does not use motion interpolation or compensation in any way. There is nothing to mod.

chibijoshie · 24th March 2013, 05:39

I was wondering if anyone had thought of modding animeIVTC to use QTGMC. I'm asking because the script uses a modded version of TempGaussMC() (precision=3 in mode=1/4, bbob=4 in mode=2, cbob=4 in mod=3), and since QTGMC is (perceived to be) better than TGMC, it was just a floating thought.

wqcr · 25th March 2013, 21:54

Thanks for your great script, -Vit- . For the last two years I relied on MCBob, which was great but extremely slow, about 1.5FPS. "Placebo" setting gives me around 3FPS which is blazingly fast for me.

Unfortunately when running the script at any preset, frames just like this appear in the final video every 350 frames or so (it is rather infrequent):

Source: DV Video PAL 720x576@25, encoded with Cedocida DV Codec through Virtualdub, fast recompress
System: Windows XP SP3

Any suggestion?

manono · 26th March 2013, 00:49

Quote:

Originally Posted by wqcr

Any suggestion?

That can sometimes be indicative of memory going bad. So you could run memtest86 or other memory testing programs to check.

Are you overclocked, by chance?

Is QTGMC the only thing in the script or are there also other complex filters?

wqcr · 26th March 2013, 09:40

Quote:

Originally Posted by manono

That can sometimes be indicative of memory going bad. So you could run memtest86 or other memory testing programs to check.

Are you overclocked, by chance?

Is QTGMC the only thing in the script or are there also other complex filters?

Memtest went fine, no errors at all. CPU is running on x10 default multiplier (C2D E7300) and no other processing apart from deinterlacing is performed.
I forgot to mention that I am using MT Avisynth 2.6 and multithread script. There were no such problems with MCBob so I guess it has something to do with the dlls.

Groucho2004 · 26th March 2013, 10:33

Quote:

Originally Posted by wqcr

Memtest went fine, no errors at all. CPU is running on x10 default multiplier (C2D E7300) and no other processing apart from deinterlacing is performed.
I forgot to mention that I am using MT Avisynth 2.6 and multithread script. There were no such problems with MCBob so I guess it has something to do with the dlls.

In order to track this down you have to start with a reasonably stable baseline:
- "official" Avisynth 2.6 Alpha 4
- Plugin package recommended by -Vit- in the first post of this thread

As a side note, I doubt that you'll get a significant performance boost with multi-threaded Avisynth on your dual core CPU. You have to consider that the encoder also needs CPU cycles. Check the speed of the entire decoding/filtering/encoding chain.

Poutnik · 26th March 2013, 13:34

Quote:

Originally Posted by Groucho2004

As a side note, I doubt that you'll get a significant performance boost with multi-threaded Avisynth on your dual core CPU. You have to consider that the encoder also needs CPU cycles. Check the speed of the entire decoding/filtering/encoding chain.

For scripts much faster than X264 ( like LRemoveDust script ) encoding with MT does not makes much sense.
But for scripts much slower MT makes big sense. Demanding scripts like MCTD or QTGMC takes much more CPU than X264.

E.g. on my core2duo E4700 2.6GHz and Vits MT AVS based on AV 2.6b3,
MCTD script takes 75-80% CPU versus 20-25% CPU for X264 for Slower preset.
I do not remember FPS gain when I compared it, but I guess it was like 1.6-1.8 times.

I think even for comparable speed of script and X264 MT can make sense, as the whole chain can be less affected by one core business by other processes. But this is rather guessing, I am not familiar with windows CPU respources policy.

I recommend modded plugin set by vit at the first page.

Boulder · 26th March 2013, 18:09

Quote:

Originally Posted by wqcr

Unfortunately when running the script at any preset, frames just like this appear in the final video every 350 frames or so

Could you post your full script? That one looks like incorrectly decoded DV, in fact it looks the same as my tapes if the tape or the head on the recorder is dirty.

wqcr · 26th March 2013, 19:34

I've figured it out. It was the multithread script and Avisynth MT. So for now I switched to Avisynth 2.58 and single-core operation, yet it is possible for me to run two instances of VDub, each one processing one-half of the footage thus occupying both cores. The result is much higher framerate than with multithread solution (4.2FPS vs. 2.9FPS) plus no artefacts

16th March 2013, 14:32	#1462 \| Link
Stormborec Registered User Join Date: Feb 2013 Posts: 161	Just two questions: 1) Is the sharpening function in qtgmc script comparable with lsfmod? At first - there is no supersample in qtgmc - is it a good idea to disable sharpening -> smode=0 ; then upscale and then to use lsfmod? 2) Is it a good idea to upscale first, if using inputtype=1? It seems, it's preserving more fine details ... Last edited by Stormborec; 16th March 2013 at 14:34.

17th March 2013, 02:14	#1464 \| Link
osgZach Registered User Join Date: Feb 2009 Location: USA Posts: 676	I'm not really tweaking anything but speed presets, and I really like the results I am getting on a particular source. The problem is I am getting line artifacts from previous frames showing up very faintly in some parts. It happens in both static and motion scenes, and I'm not sure what I can tweak to try and get rid of it.. It's not always entire outlines of entire objects per-say, just some faint remnants that show up on certain parts of the new frame, in the place where they were in the old frame.? I guess some kind of blending or residual image. I believe it is also the case that immediately following these frames with artifacts, a much cleaner frame which is almost always an exact duplicate seems to appear in the stream as well. edit: I think the source may have field blends, no idea if its a conversion though. But I'm a little rusty on procedures used to identify field blending, however. But it does appear artifacts showing up are coming from both C + P frame and possibly N frames too. There do seem to be progessive frames sandwhiched between most if not all of them as well. You can usually tell something is up because the frame image overall does not but there are subtle shifts resulting in some lines looking thinner/thicker and stuff like that.. No idea what to make of this...will post a sample later when I have time, I guess. But any ideas to try in the mean time would be nice. Last edited by osgZach; 17th March 2013 at 11:18.

21st March 2013, 09:32	#1465 \| Link
kypec User of free A/V tools Join Date: Jul 2006 Location: SK Posts: 826	Calling for help on deblending difficult scene Dear fellows, I've been using QTGMC with great results so far but recently I spotted very annoying artifacts during normal playback of my encoded footage (PAL 25fps). The source sample (AVS + D2V + M2V) is available for download here (43MB). Download also fixed AVS + D2V and replace original files that point to non-existent VOB file. Four consecutive frames that demonstrate my problem below - please note how every other frame is blended, this is actually causing very ugly stutter or judder(?) during normal watching of encoded deinterlaced progressive video @ 25fps. I've been trying to fight this with few ShutterBlur, ShutterAngleSrc, ShutterAngleOut parameter combinations but achieved no visible improvement apart from slightly blurred frames that were originally non-blended. The blended ones remain intact Can anyone recommend any possible solution to this? Whatever advanced plugin/script that could help? I don't mind if it's slow as long as it doesn't damage otherwise clean parts of deinterlaced video where QTGMC provides excellent output (watch the rest of the sample opening scene). Only these panning scenes are irritating me. Perhaps I'll try to encode source as x264 interlaced and see how my TV deinterlaces this on playback if no viable solution via Avisynth can be found... P.S. I prefer to stick with 25fps output (single-rate) if possible because I'm afraid that double-rate (50fps) would be too big for final x264 encode. Last edited by kypec; 21st March 2013 at 10:43.

21st March 2013, 09:54	#1467 \| Link
Didée Registered User Join Date: Apr 2002 Location: Germany Posts: 5,391	The sample in the ZIP is a fieldblended PAL-to-NTSC conversion. To get progressive 25fps again, use Srestore after bobbing to 60fps . __________________ - We´re at the beginning of the end of mankind´s childhood - My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!)

21st March 2013, 11:23	#1470 \| Link
Didée Registered User Join Date: Apr 2002 Location: Germany Posts: 5,391	Oops sorry, it is indeed 25i. I just looked at the mediplayer's quickinfo field, and it falsely reported the source framerate being 29.97. Important point is that QTGMC is not at fault, because the source already contains all those blendings that you issued. Anyhoo, the strategy stays the same. Just bob it, und stuff it into Srestore. --EDIT-- However, it seems that the source requires to be restored to 25fps directly, instead of Srestore's default of 23976 for PAL input. source bob Srestore(frate=25.0) In fact, the scheme of blending is quite similar to the one that is typical for PAL-to-NTSC conversions. (Oversimplified: there are not quite as much blended fields, but a lot more quasi-duplicated fields.) This is also why I "trusted" the false report of the source being 29.97 fps: because the blending pattern looks like in blended NTSC. __________________ - We´re at the beginning of the end of mankind´s childhood - My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!) Last edited by Didée; 21st March 2013 at 11:39.

21st March 2013, 09:51	#1466 \| Link
osgZach Registered User Join Date: Feb 2009 Location: USA Posts: 676	fyi, your d2v was pointing to the wrong file.. I just reindexed it. I am guessing the original source was 23.976 NTSC and it went through a PAL conversion, so maybe you need to consult QTGMC's documentation for how to deal with NTSC-> PAL conversions. I don't know much about dealing with them myself. But there are also tools like Srestore to look in to as well.

21st March 2013, 21:08	#1471 \| Link
bcn_246 Registered User Join Date: Nov 2005 Location: UK Posts: 117	Can anybody mod srestore to use SVP? tia

21st March 2013, 21:14	#1472 \| Link
Didée Registered User Join Date: Apr 2002 Location: Germany Posts: 5,391	Why, how, what? Srestore does not use motion interpolation or compensation in any way. There is nothing to mod. __________________ - We´re at the beginning of the end of mankind´s childhood - My little flickr gallery. (Yes indeed, I do have hobbies other than digital video!)

24th March 2013, 05:39	#1473 \| Link
chibijoshie Registered User Join Date: Feb 2013 Posts: 1	I was wondering if anyone had thought of modding animeIVTC to use QTGMC. I'm asking because the script uses a modded version of TempGaussMC() (precision=3 in mode=1/4, bbob=4 in mode=2, cbob=4 in mod=3), and since QTGMC is (perceived to be) better than TGMC, it was just a floating thought.

25th March 2013, 21:54	#1474 \| Link
wqcr Registered User Join Date: Sep 2007 Posts: 24	Thanks for your great script, -Vit- . For the last two years I relied on MCBob, which was great but extremely slow, about 1.5FPS. "Placebo" setting gives me around 3FPS which is blazingly fast for me. Unfortunately when running the script at any preset, frames just like this appear in the final video every 350 frames or so (it is rather infrequent): Source: DV Video PAL 720x576@25, encoded with Cedocida DV Codec through Virtualdub, fast recompress System: Windows XP SP3 Any suggestion?

26th March 2013, 19:34	#1480 \| Link
wqcr Registered User Join Date: Sep 2007 Posts: 24	I've figured it out. It was the multithread script and Avisynth MT. So for now I switched to Avisynth 2.58 and single-core operation, yet it is possible for me to run two instances of VDub, each one processing one-half of the footage thus occupying both cores. The result is much higher framerate than with multithread solution (4.2FPS vs. 2.9FPS) plus no artefacts