TIVTC settings

[radusound]

Still remain some artefacts after inverse telecine default and srestore ( hybrid ) , is something i can do to improve this ?


[Attachment 58110 - Click to enlarge]

[hintlou]

source?for example

[radusound]

Example.

[Sharc]

Try

TFM().TDecimate()
QTGMC(InputType=3)  #cleaning up

[radusound]

Thank you but i don't know to insert this code in hybrid.

[davexnet]

Thank you but i don't know to insert this code in hybrid.

Then don't use Hybrid. Create a simple script manually and encode it in Virtualdub2

[hello_hello]

SRestore is for removing blending, when field blending is used to convert the frame rate, usually between NTSC and PAL. There's no need for it in this case.
TFM is leaving behind some combing (mostly around scene changes), but changing a couple of it's defaults fixes most of it, and if you're very fussy, any fine combing that remains can be cleaned up with the Vinverse plugin.
http://avisynth.nl/index.php/Vinverse

I'm not familiar enough with Hybrid to tell you how to change those things in a script, although I'd be surprised if it didn't include Vinverse.

I'd also include some sort of noise removal, and even though it's a de-interlacer, QTGMC is quite good at cleaning up DVD video if you use it in progressive mode. There's two samples below. One without QTGMC denoising and one with it. I'm pretty sure Hybrid includes QTGMC, but once again I don't know how to change Hybrid's options. I also cropped and resized the samples using the CropResize script in my signature.

mpeg2source("D:\01x04 Emancipationtester (1)muxer.d2v")
TFM(Metric=1, MI=40)
TDecimate()
Vinverse()
QTGMC(InputType=1, EzDenoise=1.5) # only added for sample 2
CropResize(832,468, 4,0,-4,0, InDAR=16.0/9.0)

[radusound]

Thank you Hello_Hello sample 2 is what i looking for , now i have to see how translate this in hybrid ... i'm not very familiar with scrips...
TFM settings in hybrid is different , i have mode and pp ...

[hello_hello]

Hybrid must save the scripts it uses for encoding, so if anything you could probably open and modify them manually, but doesn't Hybrid have a section under a tab somewhere for manually modifying scripts? I can't remember.

[radusound]

I am very close to sample2 but i have a kind of colour ghosting...


[Attachment 58138 - Click to enlarge]

[hello_hello]

I noticed that and it's the chroma from one frame/field bleeding into another, for want of a better description. I don't know what causes it but I've seen it before. I don't think there's an automated way to fix it but someone else may know better. It only happens on some scene changes, and TFM/TDecimate is probably removing some of the bad frames as part of the IVTC process, but it's built into the actual fields.

If you compare the two screenshots below you can see the chroma from adjacent fields is somehow bleeding into each other (the bottom field in screenshot #1 and the top field in screenshot #2). The screenshots are from your sample, no filtering applied. Just the script below.

SeparateFields()
A = SelectEven()
B = SelectOdd()
StackVertical(A, B)


[Attachment 58141 - Click to enlarge]


[Attachment 58142 - Click to enlarge]

I can have a play with TFM settings to see if there's a way to prevent it from using those fields when field matching, or make it less likely, but depending on where they are in the telecine pattern it mightn't be possible. I probably won't get a chance to play around until tomorrow now, but I'll report back when I have, or maybe someone else will come along with a clever idea in the mean-time.

[radusound]

I try Srestore and seems to help...What do you think ?

[hello_hello]

Looking through the sample I uploaded I can only see one frame with chroma bleeding. The IVTC process seems to have removed the others. It's gone from your sample though, so if SRestore is fixing it, then whatever works, but can you post the script you used for that? I'm just curious to see what it is, as I was expecting problems such as the wrong frame rate due to using SRestore (because it's easy to go wrong if you don't know how to use it correctly), but it seems okay. There's still some minor combing but it's not as bad as your original screenshot.

[jagabo]

I saw 4 shot changes with chroma blending. Depending on which field TFM() starts with (field=0, or field=1) you may get one or not at those shot changes. Using TFM().SRestore(frate=23.976) instead of TFM().TDecimate() can help eliminate them. You can also use QTGMC().SRestore(frate=23.976). Another trick I sometimes use is:

# smart bob a video using TFM
function TFMBob(clip v, int "pp")
{
    pp = default(pp, 6)
    v.GetParity() ? Interleave(TFM(v, field=1, pp=pp), TFM(v, field=0, pp=pp)) : Interleave(TFM(v, field=0, pp=pp), TFM(v, field=1, pp=pp))
}

WhateverSource()
TFMBob()
SRestore(frate=x)

Although this clip was mostly simple telecined film (with a few artifacts from slowing or speeding shots, of some other mishandling) you're going to find many other parts of that series are not. Some shots will be telecined film with 30p or 30i effects overlaid, etc.

[hello_hello]

I saw 4 shot changes with chroma blending.

If you're referring to the sample I uploaded, yes you're right. I obviously looked at it in too much of a hurry yesterday.

[hello_hello]

I hadn't realised until now, but TFM has two chroma options. One to determine whether chroma is used for field marching (enabled by default), and the other to determine whether it's used for combing detection. Enabling the latter option fixed three of the bad frames, but made the fourth one worse. I don't know why. Maybe it's some sort of anomaly, or maybe not, but for the attached sample I dumped that frame manually as I couldn't find a way to get TIVTC to remove it automatically.

TFM(Metric=1, MI=40, Chroma=true)
TDecimate()
Vinverse()
QTGMC(InputType=1, EzDenoise=1.5)
CropResize(832,468, 4,0,-4,0, InDAR=16.0/9.0)
Trim(0,655) ++ Trim(655,655) ++ Trim(657,0)

[Sharc]

This works here:

Telecide().TDecimate() #no color blending; skips/repeats frame instead
QTGMC(InputType=3)  #cleaning most of the residual combing

[radusound]

Hello_hello this settings i use


[Attachment 58152 - Click to enlarge]

[hello_hello]

This works here:

Telecide().TDecimate() #no color blending; skips/repeats frame instead
QTGMC(InputType=3)  #cleaning most of the residual combing

It never occurred to me to try a different field matcher, but based on that I went back to TFM again and discovered I was causing it to keep the fourth blended frame. I don't quite understand why increasing the combing sensitivity (reducing the MI or cthresh value) causes TFM to de-interlace a frame rather than use a different match, but it seems that's what was happening. For the first three problem frames, all that was needed was to enable chroma for combing detection, but for the fourth blended frame it seems the minimum MI value must be 54.

TFM(Metric=1, MI=53, chroma=true, display=true)


[Attachment 58150 - Click to enlarge]

TFM(Metric=1, MI=54, chroma=true, display=true)


[Attachment 58151 - Click to enlarge]

The metric used doesn't seem to make much difference in this case, although the metric from Telecide (Metric=1 for TFM) does seem to be better at detecting fine combing at times, so with these settings TFM should produce much the same result as Telecide.

TFM(Metric=1, MI=55, Chroma=true)
TDecimate()
Vinverse()
QTGMC(InputType=1, EzDenoise=1.5)
CropResize(832,468, 4,0,-4,0, InDAR=16.0/9.0)

[radusound]

Thank you Sharc for sugestion , i try with qtgmc input 3 and input 2 and seems to clean but with Tfm , i try with telecide.
Still don't know how modifying script...

[hello_hello]

Hello_hello this settings i use

If you use the TFM settings from my previous post, or use Telecide instead of TFM, you shouldn't need SRestore.
http://avisynth.nl/index.php/Decomb
I'd be thinking Vinverse should come before QTGMC if you're going to use Vinverse, although I haven't tried it the other way around.

For QTGMC, Input=1, 2 and 3 are all progressive modes. Input=2 throws away every second scanline and interpolates new ones. Input=3 does the same but for the alternate scanlines.
Input=1 doesn't throw away scanlines so it'll generally retain a little more fine detail, which is why I'd be inclined to go with Vinverse().QTGMC(Input=1, EzDenoise=1.5), but I haven't tried Input 2 or 3 in this case, so whatever works better.

Maybe start a new thread for the Hybrid script editing question. The Hybrid author posts here but he or any other Hybrid users mightn't look at this thread.

[Sharc]

Thank you Sharc for sugestion , i try with qtgmc input 3 and input 2 and seems to clean but with Tfm , i try with telecide.
Still don't know how modifying script...

You can probably insert this simple script under Filtering->Avisynth->Custom in Hybrid.
Or ask in the Hybrid thread or in Selur's Forum how to do it. Sometimes GUIs are just a bit too clumsy for simple tasks

[radusound]

I try with telecide , decimate , vinverse, qtgmc and works , the ghosting is gone , but i have some artefacts ( same with first picture from first post ) with telecide default , where is m=1 and mi=55 in telecide settings , there is guide , gtresh , btresh , dtresh , ntresh , vtresh ... ?


[Attachment 58157 - Click to enlarge]

[hello_hello]

I don't think there is a Metric argument for Telecide. You don't need it anyway as Metric=1 for TFM just tells TFM to use the same metric for combing detection as Telecide.

For TFM, MI is one of the arguments for adjusting the sensitivity of the combing detection. For Telecide, I think vthresh is the equivalent, but you may have to adjust it in combination with dthresh. I haven't used Telecide much so I'd simply be experimenting myself. I suspect though, if you set vthresh too low it might cause Telecide to keep some of the blended frames as setting MI too low does for TFM, so be careful there. Reducing vthresh should help Telecide detect more of the combing (as reducing MI does for TFM), but I suspect the result will be much the same. I also suspect the only reason Telecide does better with the blended fields by default is because chroma is enabled for combing detection by default, whereas for TFM it isn't. I haven't tried disabling chroma for Telecide to test that theory though.

[radusound]

I try with vtresh=20(50 default) and seems to help a bit and dtresh=7(default). For dtresh i just don't know which value to try....

[Sharc]

I try with vtresh=20(50 default) and seems to help a bit and dtresh=7(default). For dtresh i just don't know which value to try....

For more information see the manual "DecombReferenceManual.html" which is included in the decomb package. For the tweaking you may set the parameter 'show=true'.
I think the defaults are near optimum though.

[killerteengohan]

I have almost always been able to fix or reduce visibility of those with santiag(strv=0, nns=4)

I also sometimes choose to use QTGMC to stop those from happening instead of using TIVTC, but the santiag usually does the trick.

[radusound]

Thank you for advise , i will try. I notice that in some scenes the motion it's not smooth , if i don't decimate to 23.976 and leave it to 29.97 will be better ?

[hello_hello]

No, it'll be worse, or just as bad. If you don't decimate the film sections, every 5th frame will be a duplicate.

TDecimate has hybrid modes for handling content that's a mixture of telecined (film) and interlaced (video). It can either duplicate frames in the film sections using frame blending to make the whole output 29.97fps, or it can use frame blending while removing frames from the interlaced sections to make the whole output 23.976fps. Neither is ideal, but you'd probably want to use the hybrid mode for converting the video sections to 23.976fps, given it'd be mostly film.

Alternatively, TIVTC can be used for variable frame rate encoding. It's a 2 pass process (the first pass creates metric files and it's fairly quick) and that way the film sections stay at 23.976fps and the video sections stay at 29.97fps. There's details in the TDecimate help file but post back if you need help with it.

Of course everything above assumes the sections that don't look smooth are video or interlaced at 29.97fps and the decimation is making them jerky. If that's not the case it might help to upload a sample of a problem section.

[davexnet]

This thread has ended up exactly the same as the previous thread that discussed the exact same clip.

Are there some scenes in the DVD where 29.97 interlaced was used Vs. 24 fps with pulldown ?