CTools

[real.finder]

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

that not work for vfr case with 60i :) and I need "2-way match" which mean (p/c) for tff or (c/n) for bff

[Ceppo]

that not work for vfr case with 60i and I need "2-way match" which mean (p/c) for tff or (c/n) for bff
Let me explain, so that we are sure we are comparing apple with apple, since I can't follow you when you say tff and bff.
The parity is checked internally, so you will get the correct n match whatever your source is tff or bff. If you use AssumeTFF on a BFF clip (or the inverse) you will get a TFM u match. So I don't understand why on TFF you get a p match and on BFF you get a n match. The use of that, what it is?
U,B,P matches are useless. Here why:
P match: it is useful only on the last telecined frame before a post telecine cut. If you match with the previous you will eliminate the telecined frame but you will end up with a duplicate. So it is better to deinterlace and save the field without a match to retain the original content of the source. Another use is on the first frame of a post telecine cut, if the first frame is telecined the n match will end up giving you a duplicate of the following frame. So you lose the field without match forever.
B match: makes no sense in any way possible.
U match: it may be usefull in one single case. You get a frame with the correct field order, BUT the field are swapped cronologically. A u match will fix that. But to check for it is slow for not much gain so I was thinking of making a function to use before CTelecine to fix the parity of the frame in the case a u match is what you need, so that CTelecine can get the u match by doing his standard n match.

So if I'm missing something please explain your point.

[Ceppo]

mua gmail edu hÃ*ng tốt giá rẻ có dịch vụ giao hÃ*ng thu tiá»�n táº*n nÆ¡i đảm bảo cho khách hÃ*ng. phụ bạn má»™t tay chúc bạn ngÃ*y phát triển
Or is another language or my browser don't support the characters. So I can't read.

[ChaosKing]

It's a spam post! (not the first time)

[Ceppo]

It's a spam post! (not the first time)
I see, never seen that in my life. Thanks for clarifying.

[kedautinh12]

mua gmail edu hÃ*ng tốt giá rẻ có dịch vụ giao hÃ*ng thu tiá»�n táº*n nÆ¡i đảm bảo cho khách hÃ*ng. phụ bạn má»™t tay chúc bạn ngÃ*y phát triển
Or is another language or my browser don't support the characters. So I can't read.

I knew that language. It's Vietnamese language. You can read it if forum support Vietnamese :D

[real.finder]

Let me explain, so that we are sure we are comparing apple with apple, since I can't follow you when you say tff and bff.
The parity is checked internally, so you will get the correct n match whatever your source is tff or bff. If you use AssumeTFF on a BFF clip (or the inverse) you will get a TFM u match. So I don't understand why on TFF you get a p match and on BFF you get a n match. The use of that, what it is?
U,B,P matches are useless. Here why:
P match: it is useful only on the last telecined frame before a post telecine cut. If you match with the previous you will eliminate the telecined frame but you will end up with a duplicate. So it is better to deinterlace and save the field without a match to retain the original content of the source. Another use is on the first frame of a post telecine cut, if the first frame is telecined the n match will end up giving you a duplicate of the following frame. So you lose the field without match forever.
B match: makes no sense in any way possible.
U match: it may be usefull in one single case. You get a frame with the correct field order, BUT the field are swapped cronologically. A u match will fix that. But to check for it is slow for not much gain so I was thinking of making a function to use before CTelecine to fix the parity of the frame in the case a u match is what you need, so that CTelecine can get the u match by doing his standard n match.

So if I'm missing something please explain your point.

ok, I forget that in bob case we need both (p/c) and (c/n) depending on which field we are in now, so your "CTelecineNew(bob=true).SelectEven() is p match only." is the answer I wanted

anyway, about the mouth problem, I note CTelecineNew(mode=4) kinda worst than tfm(pp=0), they should be same I think

edit: and what about adding optional clip parameter in CTelecineNew to make it do only match math from (and output from original of course), so we can use prefilterd clip with some denoise that fits the source we work on? also with it the nt will be 0 by default

[Ceppo]

anyway, about the mouth problem, I note CTelecineNew(mode=4) kinda worst than tfm(pp=0), they should be same I think
I will look into that. EDIT: Did you try sse=false? Is true by default, TFM default should be false. EDIT: you should match the nt setting as well.

edit: and what about adding optional clip parameter in CTelecineNew to make it do only match math from (and output from original of course), so we can use prefilterd clip with some denoise that fits the source we work on? also with it the nt will be 0 by default

Yes, I can add a dClip for that. Now I'm working on reworking the old functions and adding colorspaces. So I will need some days before releasing. About the nt default it must first tested, since noise is not the only source of error. The denoise might miss things like compression artefacts or whatever. Also the denoise should* be field based so it would not be very accurate.

[real.finder]

CTelecineNew(mode=4, sse=false) did it, thanks

About the nt default it must first tested, since noise is not the only source of error. The denoise might miss things like compression artefacts or whatever. Also the denoise should* be field based so it would not be very accurate.

indeed, who will using this dClip should already advanced enough to know these things

you should match the nt setting as well

how? tfm don't has nt

[Ceppo]

Strange, since FieldDiff has it, so why TFM doesn't have it :eek:

[real.finder]

the best what I got now for the mouth problem
bob1=stgmc(useEdiExt=2,EdiExt=sanimebob(useBWDIF=true, FastNnediHBD=true),Sharpness=0,tr2=2,ESearchP=true, Refinemotion=true, fastma=true)
istff=GetParity()
bob2=Interleave(TFM(field=istff ? 1 : 0, micmatching=0, mode=0, clip2=bob1.selecteven()), TFM(field=istff ? 0 : 1, micmatching=0, mode=0, clip2=bob1.selectodd()))
#~ CTelecineNew(bob=true,write=true,thr60i=0.2,nt=5,ntn=4,mode=4, sse=false)
CTelecineNew(bob=true,thr60i=0.2,inputTxt=true,nt=5,ntn=4,mode=4, sse=false)
CPostProcessing(0.1,0.2,isbob=true,edeint=bob1,edeint2=bob2)
mask = Sharpen(0,0.5).blur(1.58,0).Sharpen(0,0.2).blur(1.58,0).Sharpen(0,0.1).blur(1.58,0).CombMask (http://avisynth.nl/index.php/MTCombMask)(u=3,v=3)
last.MaskedMerge2 (http://avisynth.nl/index.php/CombMask)(bob2, mask,mi=30,blockx=8,blocky=8)
CDecimateNew(isBob=true,error=false)

[Ceppo]

What's the difference between TFM's bob and CTelecine's bob?

[real.finder]

What's the difference between TFM's bob and CTelecine's bob?

TFM cant do bob by itself, but anyway, as you can see, tfm got clip2, I know I can do CTelecineNew.CPostProcessing to get similar thing but not worth it since it's just a test

[Ceppo]

I want to give an update. I added colorspace support and rewritten most filters, but at the moment gas price increased and I'm cold. So I'm taking it slowly. You won't have to wait a whole month, but is not going to come out tomorrow. Also, I will upload everything to Github before releasing.

[Selur]

Any plans for a Vapoursynth version too?

[Ceppo]

Any plans for a Vapoursynth version too?
I don't use VapourSynth, BUT when I'm done with HBD and optimizations I might do it, but not sure.

[Ceppo]

v1.2.0 is out, please read the readme file.

Also, I adopted a dog that was abandoned and she cries every night and I can't sleep, and I can't figure out out to release the draft from GitHub, since my head hurts if someone can help :D

Also, there might be some semantic bug going on since I'm out of shape and time. Please, feedback is appreciated.

[ChaosKing]

and I can't figure out out to release the draft from GitHub, since my head hurts if someone can help :D


You just click on "Create a new release" under releases (right sidebar)

There is also no CTools repo!? Probably set to private? https://docs.github.com/en/repositories/managing-your-repositorys-settings-and-features/managing-repository-settings/setting-repository-visibility#changing-a-repositorys-visibility

[Ceppo]

Thanks a lot, I made it public :)
About the release, I did it but it is stuck forever in draft mode and I'm not in the right mind to figure out how to change that :p

[real.finder]

thanks for the update

will play with it later since I got no time now

[Ceppo]

Thanks real.finder, I'm eager to hear your opinion.

Also, I couldn't make a release cuz I didn't put a tag, and I didn't get it cuz my brain capacity was at 1% yesterday.

[ChaosKing]

Glad it's back at 2% again :D

[Ceppo]

Truth hurts lol

[real.finder]

did some tests but I think I find a bug

CTelecine(bob=true,write=true,thr60i=0.2,nt=5,ntn=4,mode=4, sse=false, dclip=SeparateFields().Removegrain(4).Weave())

it crash avspmod with WindowsError: exception: access violation reading 0x0000000000000000 in "Run analysis pass"

[Ceppo]

Thanks,
seems to happen with bob=true, I will look into it.

[Ceppo]

Fix'd.

v1.2.1 is out.

I hope everything is alright, I still need to figure out this GitHub stuff :D

[Ceppo]

@real.finder
Any new test you did with my filter? Also, if you have some filter of yours that needs to be tested today I have not much to do. :)

[real.finder]

@real.finder
Any new test you did with my filter? Also, if you have some filter of yours that needs to be tested today I have not much to do. :)

Not yet, I am kinda busy these days :) but I will do soon (maybe today)

btw, you still didn't give your Feedback about https://forum.doom9.org/showpost.php?p=1965813&postcount=843 output :rolleyes:

[Ceppo]

QTGMC()
https://i.ibb.co/09bx3Wt/qtgmc.png
sanimebob2()
https://i.ibb.co/8cg5zpZ/sanimebob2.png

bob = QTGMC()
CTelecine(bob=true,sse=false,dClip=CFieldSofter(4))
CPostProcessing(isBob=true,edeint2=bob)

https://i.ibb.co/2t68N0N/script.png

Your script has a problem on the mouth that qtgmc doesn't have (it still gets it wrong tho), also yours seems to retain a bit more details than my script however, my script is 250fps+ while sanimebob is about 4-5fps. For the rest didn't find much difference, could you try mod it, I'm interested in the details it retains :)

[real.finder]

QTGMC()
https://i.ibb.co/09bx3Wt/qtgmc.png
sanimebob2()
https://i.ibb.co/8cg5zpZ/sanimebob2.png

bob = QTGMC()
CTelecine(bob=true,sse=false,dClip=CFieldSofter(4))
CPostProcessing(isBob=true,edeint2=bob)

https://i.ibb.co/2t68N0N/script.png

Your script has a problem on the mouth that qtgmc doesn't have (it still gets it wrong tho), also yours seems to retain a bit more details than my script however, my script is 250fps+ while sanimebob is about 4-5fps. For the rest didn't find much difference, could you try mod it, I'm interested in the details it retains :)

can I got a short sample that have this frame?

[Ceppo]

https://www.mediafire.com/file/at0mup1flbkuutb/Pokimooon.m2v/file
I gave you bit larger sample, since there is a lot of stuff going on. This DVD has a lot of strange things, if you need other samples I will upload them.

That frame is probably a B/U match, it has the fields switched in chronological order.

[real.finder]

yes that kind of source is bad! it also got chroma blend, maybe we need dClip for CPostProcessing too in this case

[real.finder]

how about this?
# canimebob v1.0
function canimebob(clip i, int "stgmc_type", val "bobpresmooth", float "Str", float "Amp", bool "TV_range", int "tr2", bool "repblend", int "nnrep", string "tfm_params", int "NNSize", int "NNeurons", int "EdiQual", int "EdiMaxD", bool "FastNnediHBD") {
sisphbd = AvsPlusVersionNumber > 2294
SBitsn = sisphbd ? i.BitsPerComponent() : 8
FastNned = Default(FastNnediHBD, SBitsn > 8)
nnrep = default (nnrep, 2)
stgmc_type = default (stgmc_type, 0)
tfm_params = default(tfm_params, "")
Str = default (Str, 1.5)
tr2 = default (tr2, stgmc_type==0 ? 2 : 5)
STGMC_lsb = sisphbd ? SBitsn > 8 ? false : true : true
STGMC_n16 = STGMC_lsb && AvsPlusVersionNumber < 3382
STGMC_lsb = STGMC_n16 ? false : STGMC_lsb
istff=GetParity(i)
ii = FastNned ? i.convertbits(8, dither=1) : i
prefiltered_i = defined(bobpresmooth) ? isclip(bobpresmooth) ? bobpresmooth : IsString(bobpresmooth) ? Eval("i." + bobpresmooth) : undefined() : undefined()
iforbob = defined(prefiltered_i) ? prefiltered_i : ii.sh_bob(0,0.5).reduceflicker(strength=2).interlaced60or50(BFF=!istff)
iforbob = !defined(prefiltered_i) ? iforbob.SeparateFields().Repair(ii.SeparateFields()).weave() : iforbob
threads = sh_GetUserGlobalIMTint()
ib = FastNned ? i.sh_bob(0,0.5) : i
pni = iforbob.sh_Padding(2,4,2,4,threads=threads).nnedi3(-2, nsize=NNSize, nns=NNeurons, qual=EdiQual,threads=threads).crop(2,4,-2,-4,true)
ei = ii.eedi3(-2, mdis=EdiMaxD,sclip=pni,threads=threads)
pni = FastNned ? ib.slimit_dif(pni.convertbits(SBitsn), thr=1, elast=1.5) : pni
ei = FastNned ? ib.slimit_dif(ei.convertbits(SBitsn), thr=1, elast=1.5) : ei
ei = nnrep > 0 ? nnrep > 1 ? ei.slimit_dif2(pni,thr=4) : ei.Repair(pni,9) : ei

cbob = i.CTelecine(bob=true,nt=5,ntn=4,mode=4, sse=false,dclip=i.CFieldSofter(4))
cbob = cbob.CPostProcessing(isBob=true,edeint2=ei,mode=3)
bobclip = stgmc_type == 0 ? i.stgmc(tr0=-2,useEdiExt=1,EdiExt=cbob,Sharpness=0, Refinemotion=true, fastma=true, lsb=STGMC_lsb, n16=STGMC_n16, TV_range=TV_range, Str=Str, Amp=Amp, tr2=tr2, EdiThreads=threads) : i.stgmc(useEdiExt=2,EdiExt=cbob,Sharpness=0,ESearchP=true, Refinemotion=true, fastma=true, lsb=STGMC_lsb, n16=STGMC_n16, TV_range=TV_range, Str=Str, Amp=Amp, tr2=tr2, EdiThreads=threads)
i.sh_Padding(0,4,0,4,threads=threads).yadifmod2(mode=3, edeint=cbob.sh_Padding(0,4,0,4,threads=threads)).crop(0,4,-0,-4,true).Vinverse(clip2=pni).slimit_dif2(bobclip,thr=4)
}
originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)
canimebob
when CPostProcessing got dClip it can be improved

edit: btw, I did update sanimebob2 too https://forum.doom9.org/showpost.php?p=1965813&postcount=843

[real.finder]

edit: and what about adding optional clip parameter in CTelecineNew to make it do only match math from (and output from original of course), so we can use prefilterd clip with some denoise that fits the source we work on? also with it the nt will be 0 by default

Yes, I can add a dClip for that. Now I'm working on reworking the old functions and adding colorspaces. So I will need some days before releasing. About the nt default it must first tested, since noise is not the only source of error. The denoise might miss things like compression artefacts or whatever. Also the denoise should* be field based so it would not be very accurate.

did some tests but I think I find a bug

CTelecine(bob=true,write=true,thr60i=0.2,nt=5,ntn=4,mode=4, sse=false, dclip=SeparateFields().Removegrain(4).Weave())

it crash avspmod with WindowsError: exception: access violation reading 0x0000000000000000 in "Run analysis pass"

Thanks,
seems to happen with bob=true, I will look into it.

Fix'd.

v1.2.1 is out.

I hope everything is alright, I still need to figure out this GitHub stuff :D

ok, I did test with

#~ CTelecine(bob=true,write=true,thr60i=0.2,nt=5,ntn=4,mode=4, sse=false,dclip=Blur(1.58,0))
CTelecine(bob=true,thr60i=0.2,inputTxt=true,nt=5,ntn=4,mode=4, sse=false)
CPostProcessing(3,6,isbob=true,edeint2=qtgmc(EdiMode="BWDIF+EEDI3",Sharpness=0, Refinemotion=true, n16=true, TV_range=true, tr2=2))
CDecimate(isBob=true,error=false)

with this https://forum.doom9.org/showthread.php?p=1964544#post1964544 (it was from Jigoku Shōjo 1st ep) and it fine in both "60i" and "false n matches" cases, dclip was very useful :) thanks for the fix btw

[Ceppo]

There are 2 problems with SAnimeBob2, 1. It has an interlaced leftover (look at the fade at the beginning of the clip I provided), 2. There is still that problem with the mouth.

About CAnimeBob, if you don't feed a QTGMC like clip to CPostProcessing it won't work well. With the default, I got 10fps and a much worst output than the script I made above. Also, it has an interlaced leftover problem as SAnimeBob2.

For the chroma blending, CFieldDeBlend should be able to fix most cases as long there is a duplicate field in the prev or next frame.

Can you elaborate on the use of dClip for CPostProcessing?


originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)
canimebob

What does this code do?

Finally, aside bob, all the parameters of CTelecine are ignored when inputTxt=true, so if you are lazy you don't need to write down :). Also I advice CTelecine default or CTelecine(sse=false,dClip=CFIeldSofter(4)), you might want to test them.

Thanks for your support! :D

[real.finder]

this

originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)

to fix chroma blending, see here https://forum.doom9.org/showthread.php?t=170364

Can you elaborate on the use of dClip for CPostProcessing?

CPostProcessing seems to have many false positive cases, that can be improved with filtered clip same as dClip in CTelecine

also, we need you here https://forum.doom9.org/showthread.php?p=1966602#post1966602 :)

[Ceppo]

CPostProcessing seems to have many false positive cases, that can be improved with filtered clip same as dClip in CTelecine
What kind of prefiltering would improve comb detection in your opinion?


originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)

Tomorrow I will try, I my dog is cousing trouble :D Is all in AnimeIVTC.avsi?

[real.finder]

What kind of prefiltering would improve comb detection in your opinion?


something like this maybe
Sharpen(0,0.5).blur(1.58,0).Sharpen(0,0.2).blur(1.58,0).Sharpen(0,0.1).blur(1.58,0) (https://forum.doom9.org/showpost.php?p=1965247&postcount=111)



originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)

Tomorrow I will try, I my dog is cousing trouble :D Is all in AnimeIVTC.avsi?

and https://github.com/realfinder/AVS-Stuff/blob/master/avs%202.5%20and%20up/Zs_RF_Shared.avsi ;)

[Ceppo]

Ok, I will add a dClip to CPostProcessing.

Also, I tried the chroma repair script, and it didn't work. Maybe I'm doing it wrong...

Script (nothing changes) (https://i.ibb.co/pKWPt3p/RFscript.png)
CFieldDeBlend() (https://i.ibb.co/m50wtqZ/CField-De-Blend.png)

[real.finder]

Ok, I will add a dClip to CPostProcessing.

Also, I tried the chroma repair script, and it didn't work. Maybe I'm doing it wrong...

Script (nothing changes) (https://i.ibb.co/pKWPt3p/RFscript.png)
CFieldDeBlend() (https://i.ibb.co/m50wtqZ/CField-De-Blend.png)

nope, you not doing it wrong, the method is not perfect in first place https://forum.doom9.org/showthread.php?p=1778933#post1778933 (until post #10) it kinda work in fast motion only and without 60i

[Ceppo]

Updated to v1.2.2 for CPP dClip.

Also I changed CResize default.

@real.finder
I think I will stick with CFieldDeBlend then.

[real.finder]

I did test with pokemon sample (frame 1570 to 2410) from you

originali = last
ffapp = originali.sh_bob(0,0.5).chroma_rep #or chroma_rep2 see dchr below
getparity(originali) ? ffapp.interlaced60or50 : ffapp.interlaced60or50(bff=true) #reinterlaced
MergeLuma(originali)
bob = qtgmc(EdiMode="BWDIF+EEDI3",Sharpness=0, Refinemotion=true, n16=true, TV_range=true, tr2=2)
CTelecine(dclip=Blur(1.58,0))
CPostProcessing(dclip=vsMSharpen().blur(1.58,0).RemoveGrain(),edeint2=bob.CFieldDeBlend())

it can be improved more, one of the problems is qtgmc don't like anime :devil:

[Ceppo]

Can you show me a frame in the source that the chroma repair can handle and CFieldDeBlend can't?


bob = qtgmc(EdiMode="BWDIF+EEDI3",Sharpness=0, Refinemotion=true, n16=true, TV_range=true, tr2=2)
CTelecine(dclip=Blur(1.58,0))
CFieldDeBlend() #CFieldDeBlend works only after fieldmatching
CPostProcessing(dclip=vsMSharpen().blur(1.58,0).RemoveGrain(),edeint2=bob) #This dClip gives me more false positive than default


Also, default CPostProcessing is meant to detect every mouth, so is very sensitive. You might want to change blksize/blkthr to get less of them, but some mouth might escape the check.

BTW, I think is good to use as fewer filters as possible, instead of vsMSharpen you might want to try CSharpen so that the script doesn't have much external dependency. As it is now you need:
1- QTGMC
2- vsMSharpen
3- Removegrain
4- BWDIF
5- EEDI3
6- CTools
Aside Removegrain/QTGMC, I don't think that the others are in everyone folders (I had to download them).

Also, I'm working on CTGMC, still alpha, when I'm done studing QTGMC we can try with that.

[real.finder]

Can you show me a frame in the source that the chroma repair can handle and CFieldDeBlend can't?
the point that I like to be safe and use all privileges from the filters, I didn't test much to know the Pros and Cons of CFieldDeBlend but anyway chroma_rep was based on srestore IIRC so I think it can not only replace Blend frame but also mix frames to get less Blend

BTW, I think is good to use as fewer filters as possible

I agree, but at the same time if I can get more quality I will mix some filters, at least this is the case that was in old sanimebob

[Ceppo]

@real.finder
Today I tried implementing my bob idea from scratch, looks there is less blending/shimmering. It lacks a lot of things, however, seems promising for Anime sources. Can I PM you the test script to hear your opinion?

[real.finder]

Can I PM you the test script to hear your opinion?

sure, why not?

[Ceppo]

I sent it (I hope), if you didn't get it, please tell me.

[mastrboy]

These CTools plugins are great, thank you.

Any chance to get a 64bit version of IsCombedT (part of TCombMask) ?

[Ceppo]

At the moment I'm full of stuff to do. Like HBD/linux/Vapursynth CTools then I have CTGMC and CBicubicResize to do, so I don't have time. Also, I never used that filter so I'm not interested in messing with it. I see that the creator is tritical and he probably used some C++ concept that I have yet to study. So you might want to make a request on the forum or something, hoping that some true programmer will help you out. :)

[mastrboy]

At the moment I'm full of stuff to do. Like HBD/linux/Vapursynth CTools then I have CTGMC and CBicubicResize to do, so I don't have time. Also, I never used that filter so I'm not interested in messing with it. I see that the creator is tritical and he probably used some C++ concept that I have yet to study. So you might want to make a request on the forum or something, hoping that some true programmer will help you out. :)

Thanks for the fast reply, no problem, I'll try to solve this some other way.