CTools - Page 5

Ceppo · 27th February 2022, 01:45

Quote:

OK. Did you mention that this also works on 60 fps video source too?

I tried it on a live 60 fps video source and it made the video smoother in my opinion (motion interpolation).

Is that the point of the filter?

Can you show me an example with the other parameters to adjust for more interpolation/smoothness?

I'm currently encoding a source, so I can't encode, so, if you could wait, I will. Also if you upload a sample, I will tweak it for that sample so that I can better show you.

EDIT:
EXAMPLES
Messed a bit with mvtools. Came up with that to reduce artefacts.
MCompensate clip:

Code:

AddBorders(0,0,1280%64,720%64)
src = last
super = MSuper()
backvec1 = MAnalyse(super,blksize=4,overlap=2,isb=true,delta=1)
backvec2 = MAnalyse(super,blksize=8,overlap=4,isb=true,delta=1)
backvec3 = MAnalyse(super,blksize=16,overlap=8,isb=true,delta=1)
backvec4 = MAnalyse(super,blksize=32,overlap=16,isb=true,delta=1)
backvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=true,delta=1)
iClip1 = MCompensate(super,backvec1,time=50)
iClip2 = MCompensate(super,backvec2,time=50)
iClip3 = MCompensate(super,backvec3,time=50)
iClip4 = MCompensate(super,backvec4,time=50)
iClip5 = MCompensate(super,backvec5,time=50)

ScriptClip("""
AverageLuma(MMask(backvec1,kind=1,ml=255)) < 1 ? iClip1 : \
AverageLuma(MMask(backvec2,kind=1,ml=255)) < 1 ? iClip2 : \
AverageLuma(MMask(backvec3,kind=1,ml=255)) < 1 ? iClip3 : \
AverageLuma(MMask(backvec4,kind=1,ml=255)) < 1 ? iClip4 : iClip5
""")

CInterpolate(src,iClip=last)
Crop(0,0,-(1280%64),-(720%64))

MFlowInter clip (slow)

Code:

AddBorders(0,0,1280%64,720%64)
src = last
super = MSuper()
backvec1 = MAnalyse(super,blksize=4,overlap=2,isb=true,delta=1)
forwvec1 = MAnalyse(super,blksize=4,overlap=2,isb=false,delta=1)
backvec2 = MAnalyse(super,blksize=8,overlap=4,isb=true,delta=1)
forwvec2 = MAnalyse(super,blksize=8,overlap=4,isb=false,delta=1)
backvec3 = MAnalyse(super,blksize=16,overlap=8,isb=true,delta=1)
forwvec3 = MAnalyse(super,blksize=16,overlap=8,isb=false,delta=1)
backvec4 = MAnalyse(super,blksize=32,overlap=16,isb=true,delta=1)
forwvec4 = MAnalyse(super,blksize=32,overlap=16,isb=false,delta=1)
backvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=true,delta=1)
forwvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=false,delta=1)
iClip1 = MFlowInter(super,backvec1,forwvec1,time=50,blend=false)
iClip2 = MFlowInter(super,backvec2,forwvec2,time=50,blend=false)
iClip3 = MFlowInter(super,backvec3,forwvec3,time=50,blend=false)
iClip4 = MFlowInter(super,backvec4,forwvec4,time=50,blend=false)
iClip5 = MFlowInter(super,backvec5,forwvec5,time=50,blend=false)

ScriptClip("""
AverageLuma(MMask(backvec1,kind=1,ml=255)) < 1 ? iClip1 : \
AverageLuma(MMask(backvec2,kind=1,ml=255)) < 1 ? iClip2 : \
AverageLuma(MMask(backvec3,kind=1,ml=255)) < 1 ? iClip3 : \
AverageLuma(MMask(backvec4,kind=1,ml=255)) < 1 ? iClip4 : iClip5
""")

CInterpolate(src,iClip=last)
Crop(0,0,-(1280%64),-(720%64))

@ 24fps the artefacts are easy to spot, but it should not be so if you use it on a bobbed clip @60fps.

real.finder · 27th February 2022, 18:15

Quote:

Originally Posted by Ceppo

Will do. But what do you think about it?

I can't say unless I test

Quote:

Also can you link me the post with your filters? I can't find them.

in my forum signature blow

gispos · 27th February 2022, 18:54

Quote:

Originally Posted by Ceppo

Will do. But what do you think about it?

For me it's 3 FPS (16%) slower on full HD. It should also be renamed, ex_QTGMC or similar.
@real.finder please do not completely abandon support for the 'original' QTGMC!

Ceppo · 27th February 2022, 23:18

Quote:

For me it's 3 FPS (16%) slower on full HD. It should also be renamed, ex_QTGMC or similar.

Did you see some improvement like less ghosting or something? Once I have done with the plugin I might look into it.

Quote:

in my forum signature blow

I checked there, but I can't find the 2 filters. I'm missing something...

real.finder · 27th February 2022, 23:59

sanimebob is part of my animeivtc mod, stgmc and any new things https://github.com/realfinder/AVS-Stuff as the big note said there

Ceppo · 28th February 2022, 01:26

Mystery solved!

real.finder · 28th February 2022, 16:47

Quote:

Originally Posted by Ceppo

v1.0.9
- Added mode=3 and mode=4 to CTelecineNew.

what about adding mode 5 using IT method https://github.com/HomeOfVapourSynth...VapourSynth-IT ?

also the tfm method has slow parameter

Quote:

slow -

Sets which field matching fuction is used. Possible settings:

0 - normal (should have the worst accuracy)
1 - slower
2 - slowest (should have the best accuracy)

This is basically a quality vs. speed option (the slower the function the more
accurate it should be). Prior to v0.9.9.1 of tfm this was a boolean value, true
from previous versions corresponds to a current value of 1 and false corresponds
to a current value of 0.

Default: 1 (int)

kedautinh12 · 28th February 2022, 16:59

Avs ver
http://avisynth.nl/index.php/IT

real.finder · 28th February 2022, 17:15

Quote:

Originally Posted by kedautinh12

Avs ver
http://avisynth.nl/index.php/IT

vs one based on avs one, and avs IT has no c code so it hard to read it

kedautinh12 · 28th February 2022, 17:31

Quote:

Originally Posted by real.finder

vs one based on avs one, and avs IT has no c code so it hard to read it

I think CTools is avs ver so i share avs ver of IT is acceptable

Ceppo · 1st March 2022, 01:03

Thanks guys. About mode 5, I can't read both avs and vs versions

, I might need some time. It would be easier if I knew the "concept" of that filter so that I can implement it my own way. About TFM slow parameter, when I did TelecineBob, I found it not much needed, but I might look into it again.

EDIT: mode 4 runs at 500fps to me. What's the point of making it faster and losing precision?

Ceppo · 1st March 2022, 02:12

v1.1.0
- Removed Sqrt parameter from CSharpen.
- Added mode parameter to CSharpen + new sharpening mode.
- Added CDeHalo.
- Added CAntiAliasing.
- Added ntN parameter to CTelecineNew for mode 3, 4.
- CTelecineNew default mode is now 3.

My idea for the mouth problem didn't work on anime. So it was pointless

. I had another idea for mode 3 and mode 4. Just decrease ntN parameter to get less false n matches.

gispos · 1st March 2022, 17:45

Quote:

Originally Posted by Ceppo

Did you see some improvement like less ghosting or something? Once I have done with the plugin I might look into it.

There are minimal differences in my default settings.
For me, there is no winner. In some areas STGMC is a bit more contoured and in others Dogway's QTGMC is a bit more contoured. By that I also mean ghosting or blurr removed.
But I haven't tested any other settings. You have to test it yourself, some like it that way, others like it differently.

Milardo · 2nd March 2022, 23:17

Quote:

Originally Posted by Ceppo

I'm currently encoding a source, so I can't encode, so, if you could wait, I will. Also if you upload a sample, I will tweak it for that sample so that I can better show you.

EDIT:
EXAMPLES
Messed a bit with mvtools. Came up with that to reduce artefacts.
MCompensate clip:

Code:

AddBorders(0,0,1280%64,720%64)
src = last
super = MSuper()
backvec1 = MAnalyse(super,blksize=4,overlap=2,isb=true,delta=1)
backvec2 = MAnalyse(super,blksize=8,overlap=4,isb=true,delta=1)
backvec3 = MAnalyse(super,blksize=16,overlap=8,isb=true,delta=1)
backvec4 = MAnalyse(super,blksize=32,overlap=16,isb=true,delta=1)
backvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=true,delta=1)
iClip1 = MCompensate(super,backvec1,time=50)
iClip2 = MCompensate(super,backvec2,time=50)
iClip3 = MCompensate(super,backvec3,time=50)
iClip4 = MCompensate(super,backvec4,time=50)
iClip5 = MCompensate(super,backvec5,time=50)

ScriptClip("""
AverageLuma(MMask(backvec1,kind=1,ml=255)) < 1 ? iClip1 : \
AverageLuma(MMask(backvec2,kind=1,ml=255)) < 1 ? iClip2 : \
AverageLuma(MMask(backvec3,kind=1,ml=255)) < 1 ? iClip3 : \
AverageLuma(MMask(backvec4,kind=1,ml=255)) < 1 ? iClip4 : iClip5
""")

CInterpolate(src,iClip=last)
Crop(0,0,-(1280%64),-(720%64))

MFlowInter clip (slow)

Code:

AddBorders(0,0,1280%64,720%64)
src = last
super = MSuper()
backvec1 = MAnalyse(super,blksize=4,overlap=2,isb=true,delta=1)
forwvec1 = MAnalyse(super,blksize=4,overlap=2,isb=false,delta=1)
backvec2 = MAnalyse(super,blksize=8,overlap=4,isb=true,delta=1)
forwvec2 = MAnalyse(super,blksize=8,overlap=4,isb=false,delta=1)
backvec3 = MAnalyse(super,blksize=16,overlap=8,isb=true,delta=1)
forwvec3 = MAnalyse(super,blksize=16,overlap=8,isb=false,delta=1)
backvec4 = MAnalyse(super,blksize=32,overlap=16,isb=true,delta=1)
forwvec4 = MAnalyse(super,blksize=32,overlap=16,isb=false,delta=1)
backvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=true,delta=1)
forwvec5 = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=false,delta=1)
iClip1 = MFlowInter(super,backvec1,forwvec1,time=50,blend=false)
iClip2 = MFlowInter(super,backvec2,forwvec2,time=50,blend=false)
iClip3 = MFlowInter(super,backvec3,forwvec3,time=50,blend=false)
iClip4 = MFlowInter(super,backvec4,forwvec4,time=50,blend=false)
iClip5 = MFlowInter(super,backvec5,forwvec5,time=50,blend=false)

ScriptClip("""
AverageLuma(MMask(backvec1,kind=1,ml=255)) < 1 ? iClip1 : \
AverageLuma(MMask(backvec2,kind=1,ml=255)) < 1 ? iClip2 : \
AverageLuma(MMask(backvec3,kind=1,ml=255)) < 1 ? iClip3 : \
AverageLuma(MMask(backvec4,kind=1,ml=255)) < 1 ? iClip4 : iClip5
""")

CInterpolate(src,iClip=last)
Crop(0,0,-(1280%64),-(720%64))

@ 24fps the artefacts are easy to spot, but it should not be so if you use it on a bobbed clip @60fps.

Ok. I don't really have any samples, i am testing using a live 60 fps video source capture card.

I don't know if your filter is designed to do this, but can you post an example for a source like mine, to get it as as smooth as possible, i'm doing this for realtime playback. Does adjusting any of the parameters like "thr" do anything for that?

Ceppo · 3rd March 2022, 05:12

Quote:

Ok. I don't really have any samples, i am testing using a live 60 fps video source capture card.

I don't know if your filter is designed to do this, but can you post an example for a source like mine, to get it as as smooth as possible, i'm doing this for realtime playback. Does adjusting any of the parameters like "thr" do anything for that?

I don't have a source like yours. Anyway the filter is designed for videos who have duplicate frames, a non anime source should not have duplicate unless your source is telecined and instead of dropping the duplicate you interpolate it with CInterpolate + mvtools. In other cases you don't need it. CInterpolate's parameters are all about finding duplicates, if you want more interpolation set an higher thr value so that more duplicates are detected. The interpolation job is done by mvtools, CInterpolate will just detect the duplicate.

Milardo · 3rd March 2022, 06:02

Quote:

Originally Posted by Ceppo

I don't have a source like yours. Anyway the filter is designed for videos who have duplicate frames, a non anime source should not have duplicate unless your source is telecined and instead of dropping the duplicate you interpolate it with CInterpolate + mvtools. In other cases you don't need it. CInterpolate's parameters are all about finding duplicates, if you want more interpolation set an higher thr value so that more duplicates are detected. The interpolation job is done by mvtools, CInterpolate will just detect the duplicate.

src = last
super = MSuper()
backvec = MAnalyse(super,blksize=64,overlap=32,lambda=1000*64,isb=true,delta=1)
iClip = MCompensate(super,backvec,time=50)
CInterpolate(iClip=iClip, thr=1.00)
last

Would this script be the basic example and did i get "thr=1.00" a possible value?

Ceppo · 3rd March 2022, 08:36

Use this presets. You should toy with mvtools to see what you like and what you don't like tho.

Code:

function iClip(clip src)
{
    src
    super = MSuper()
    backvec = MAnalyse(super,blksize=4,overlap=2,isb=true,delta=1)
    forwvec = MAnalyse(super,blksize=4,overlap=2,isb=false,delta=1)
    MFlowInter(src,super,backvec,forwvec,blend=false,time=50)
    return last
}

function iClip2(clip src)
{
    src
    super = MSuper()
    backvec = MAnalyse(super,blksize=64,overlap=32,isb=true,delta=1)
    forwvec = MAnalyse(super,blksize=64,overlap=32,isb=false,delta=1)
    MFlowInter(src,super,backvec,forwvec,blend=false,time=50)
    return last
}

So you do:

Code:

CInterpolate(iClip=iClip())
or
CInterpolate(iClip=iClip2())

You can use thr=1.0 if you want, sure. It will detect more frames as duplicates.
You can also set a denoise to get a better duplicate detection against noise.

Code:

CInterpolate(iClip=iClip2(),dClip=FFT3DFilter(sigma=4))

real.finder · 3rd March 2022, 22:17

Quote:

Originally Posted by Ceppo

EDIT: mode 4 runs at 500fps to me. What's the point of making it faster and losing precision?

I was thinking about the opposite

which is make it slower optionally

real.finder · 3rd March 2022, 23:14

Quote:

Originally Posted by Ceppo

My idea for the mouth problem didn't work on anime. So it was pointless

. I had another idea for mode 3 and mode 4. Just decrease ntN parameter to get less false n matches.

is there a way to not match with n in first place? like mode=0 in tfm

Quote:

0 = 2-way match (p/c)

I did try ntN=0 for https://forum.doom9.org/showthread.p...70#post1964470 but it not work well

edit: the n matching doesn't make sense for "tff: top field first" source anyway as the p matching doesn't make sense for "bff: bottom field first", maybe we need option to limit the matching for p/c for tff and p/n for bff

about idea for mouth problem I remember this was a good fix https://github.com/realfinder/AVS-St...eIVTC.avsi#L84 and https://github.com/realfinder/AVS-St...IVTC.avsi#L289

the basic idea of it is to compare the output with the bob output using https://github.com/pinterf/TIVTC/blo...D%20ME.txt#L24

you can use the concept but instead of reverse the field it will be the output of "mode 4 (TIVTC)" vs "mode 3 (Decomb)" or optionally even better like mode 4 vs mode 3 vs mode 2

by maybe make "mode -1" mean "mode 4 vs mode 3" and "mode -2" mean "mode 4 vs mode 3 vs mode 2" and "mode -3" mean "mode 4 vs mode 3 vs mode 2 vs mode 1"

Ceppo · 4th March 2022, 06:50

Quote:

is there a way to not match with n in first place? like mode=0 in tfm

CTelecineNew(bob=true).SelectEven() is p match only.

Quote:

edit: the n matching doesn't make sense for "tff: top field first" source anyway as the p matching doesn't make sense for "bff: bottom field first", maybe we need option to limit the matching for p/c for tff and p/n for bff

The parity is internally detected, if change the field order like AssumeBFF() for a TFF one. You are going to get the wrong result, I think it is b/u match from TFM. Which don't make any sense to me.

Quote:

I was thinking about the opposite which is make it slower optionally

Well, it's the slowest formula already

.

Quote:

the basic idea of it is to compare the output with the bob output using https://github.com/pinterf/TIVTC/blo...D%20ME.txt#L24

I think that's something to do after CPostProcessing, that I have yet to update with the other modes. But I will think about it.

27th February 2022, 23:59	#85 \| Link
real.finder Registered User Join Date: Jan 2012 Location: Mesopotamia Posts: 2,587	sanimebob is part of my animeivtc mod, stgmc and any new things https://github.com/realfinder/AVS-Stuff as the big note said there __________________ See My Avisynth Stuff

1st March 2022, 01:03	#91 \| Link
Ceppo Registered User Join Date: Feb 2016 Location: Nonsense land Posts: 339	Thanks guys. About mode 5, I can't read both avs and vs versions , I might need some time. It would be easier if I knew the "concept" of that filter so that I can implement it my own way. About TFM slow parameter, when I did TelecineBob, I found it not much needed, but I might look into it again. EDIT: mode 4 runs at 500fps to me. What's the point of making it faster and losing precision? Last edited by Ceppo; 1st March 2022 at 01:27.

28th February 2022, 01:26	#86 \| Link
Ceppo Registered User Join Date: Feb 2016 Location: Nonsense land Posts: 339	Mystery solved!

28th February 2022, 16:59	#88 \| Link
kedautinh12 Registered User Join Date: Jan 2018 Posts: 2,156	Avs ver http://avisynth.nl/index.php/IT

1st March 2022, 02:12	#92 \| Link
Ceppo Registered User Join Date: Feb 2016 Location: Nonsense land Posts: 339	v1.1.0 - Removed Sqrt parameter from CSharpen. - Added mode parameter to CSharpen + new sharpening mode. - Added CDeHalo. - Added CAntiAliasing. - Added ntN parameter to CTelecineNew for mode 3, 4. - CTelecineNew default mode is now 3. My idea for the mouth problem didn't work on anime. So it was pointless . I had another idea for mode 3 and mode 4. Just decrease ntN parameter to get less false n matches.