TDeint and TIVTC

[Frank62]

Originally Posted by Katie Boundary

...the two fields be from slightly different images.

That exactly is how "motion" in connection with video is "defined", at least used: The two fields contain different images. Shot at a slightly different time. So what happened from the first to the second point in time is "motion" of some or all objects in the pictures.
If you can't accept this, w_m seems to be right with his last sentence. A question of terminology.

[StainlessS]

Shot at a slightly different time.

So progressive is temporally from the same position in time [excluding time taken for left to right, top to bottom, raster scan],
and interlaced fields are from temporally and also spatially different positions in time and space.
Katie should really teach herself to code, then she could better acquaint herself with the problem, that would be an education.
Miss Boundary is real exasperating, I bet its hell in their hourse deciding whats for supper.

[FranceBB]

Originally Posted by StainlessS

Katie should really teach herself to code, then she could better acquaint herself with the problem, that would be an education.

Hear hear.
(Ok, I might have been watching the house of commons channel too much xD)

Jokes aside, she would grasp some of those things much much better if she was able to read and write code herself 'cause this way she would be able to actually test her ideas while right now she can only "trust" us with what we're telling her.

Originally Posted by StainlessS

I bet its hell in their house deciding whats for supper.

ROTFL
Think about the chat with her husband, like:

Honey, how did your day go?

- Fine, I went to work, it was a normal day, I got stuck in traffic a bit on my way back home, but anyway... What about you?

Terrible, I came up with a revolutionary idea about duplicates detection, but everyone on Doom9 is saying it can't be used as it's not that simple...

[Katie Boundary]

Originally Posted by Frank62

That exactly is how "motion" in connection with video is "defined", at least used:

No. Motion means that a thing used to be in one place in space, and now it's in another.

Originally Posted by FranceBB

Think about the chat with her husband,

I'm not likely to ever have a husband, given that I'm only attracted to women.

[FranceBB]

Originally Posted by Katie Boundary

Motion means that a thing used to be in one place in space, and now it's in another.

Let's just stick with "difference" and we're good XD
No, jokes aside, we understand what you're saying, honestly, we do, but on the other hand it's just not as easy to detect differences (motion in your case) through an algorithm, but hey, if in the future someone will try to improve TDecimate and co. further he's gonna be more than welcome, but right now, if it doesn't fulfill your / our needs with any of the modes available, I guess we're just gonna have to live with it... :')

Originally Posted by Katie Boundary

I'm not likely to ever have a husband, given that I'm only attracted to women.

Ah! Of course! Sorry, my fault, everything makes sense now eheheh.

(I'm joking of course )

Ok, I should really get back to work now, I shouldn't check stuff on Doom9 while I'm at work, but I'm... like... addicted to this forum as I end up here most of the time anyway since I use Avisynth at work on a daily basis. :P

[Katie Boundary]

Originally Posted by FranceBB

Let's just stick with "difference" and we're good XD
No, jokes aside, we understand what you're saying, honestly, we do, but on the other hand it's just not as easy to detect differences (motion in your case) through an algorithm,

It totally is. There's an unlimited number of algorithms that you can make based on the guidelines that I posted, but pretty much any algorithm based on those guidelines will be better than the options present in TFM.


Example algorithm 1:

Find average of A and E.
Find average of B and D.
If C is closer to the average of A and E than to the average of B and D, then treat C as interlaced.

Example algorithm 2:

m1 = |A-B|
m2 = |B-C|
m3 = |C-D|
m4 = |D-E|
n1 = |A-C|
n2 = |B-D|
n3 = |C-E|
If m1, m2, m3, and m4 are all bigger than n1, n2, and n3, then treat C as interlaced.

[Frank62]

Originally Posted by Katie Boundary

No. Motion means that a thing used to be in one place in space, and now it's in another.

That's correct. But I wrote: "In connection with video", where your explanation/definition never happens. We are talking of pixels, bits, information, representations of pictures, not of "things".

[SaurusX]

Anyone ever run into this problem with TFM? The field matching fails to use the correct field combination and you end up with chunky aliased edges, particularly apparent in animation lines. It never happens if the source is completely stable, but if there's gate weave from a film source then hold on to your hat. You have to manually override with the correct field combo.

https://imgsli.com/NzA0MjU

This is a small-fry example. On occasion the entire frame will be huge chunky (but not combed) scanlines.

[VoodooFX]

Originally Posted by SaurusX

This is a small-fry example.

Share a short snippet of original video and your script.

[SaurusX]

Originally Posted by VoodooFX

Share a short snippet of original video and your script.

Check out frame 32 and 33 in this clip using the following script.

MPEG2SOURCE("E:\Media\Video Editing Tools\MeGUI-2913-64\scripts\SwatKats\Tests\SwatKats Example.d2v")
TFM(slow=2,pp=4,mode=4)
TDecimate(mode=1)

https://mega.nz/file/qjpgXRqK#kS4NW-...OPjYM7B-Y-mYmE

[VoodooFX]

Originally Posted by SaurusX

Anyone ever run into this problem with TFM?

Many times, especially on animations like this.

Instead of TFM try this line (check if it fixes more than it destroys):
QTGMC(Preset="Slower").SelectOdd.TDecimate(mode=1).QTGMC( Preset="Medium", InputType=1, Sharpness=0.2)

[Katie Boundary]

Originally Posted by Frank62

I wrote: "In connection with video", where your explanation/definition never happens.

LOL, are you high? It happens all the time in video. It's the primary cause of differences between frames.

[Frank62]

Senseless.

[markfilipak]

Katie Boundary! Hahaha... The joke's on you, folks. I visualize a T.Rex eating a guy who's sitting on a toilet as a comet streaks across the sky.

Edit: Oh, I see "Miss Boundary" has a T.Rex as an avatar -- that, after enlarging it to see it well. Well, "Miss Boundary", you need a better looking/better defined avatar. People are missing the joke.

[videoh]

I think you are the only guy that missed it.

[VoodooFX]

Originally Posted by Katie Boundary

Unless it's around a scene change. Or it's a fade. Or there's literally any other change in the image that isn't caused by motion (someone turns a light on, for example).

But photons are moving from a bulb to a camera's lens...
A scene change and a fade... thermodynamics would tell that there is no motion, but quantum mechanics would disagree, I think. Anyway, a film editor moved his hand to make those effects, so literally they are caused by motion too.

[markfilipak]

Originally Posted by videoh

I think you are the only guy that missed it.

I missed it? Hahaha... Missed what? I laughed my ass off. This whole thread segment with "Miss Boundary" is the silliest thing I've read in a long time. Thanks for the laughs.

[LigH]

Well, unfortunately it distracts from the topic. Much of the "discussion of basics of interlacing" deserves a separate thread. And a lot less personal insults. After all the time and number of involved people, it is now a bit hard to separate.

[kedautinh12]

Latest ver:
https://github.com/pinterf/TIVTC/releases

[Blankmedia]

From time to time I get this error with requestlinear:

"avs+ [FLAW]: RequestLinear: internal error (frame not cached)! occurred while reading frame 0"

And this is while running the same script, like I stopped the encode for whatever reason, start it back, get the error, flush intermediate files, get the error, reboot, get the error, type this message, no more error and encode is proceding (??).

Maybe I have trouble settings the parameters, so is there bigger documentation somewhere with requestlinear.

I run it because apparently there's a random bug in mvtools where it shoots a random frames from time to time.

This is the line I use :

RequestLinear(rlim=60, clim=100, elim=40)

it seem more stable than what I used before

RequestLinear(rlim=50, clim=100, elim=25)

Edit:

I run the script through x265 patman.

And even in avsmeter it started to fail. Avsmeter doesn't say anything, it just close quickly.

Edit 2:
it is not more stable.

I don't get it.

PHP Code:

ClearAutoloadDirs()
#~ AddAutoloadDir("D:\A encoder\avs_simpsons_plugin")
AddAutoloadDir("D:\A encoder\simps_s02_60FPSv3\simps_s02_60FPSv3_plugin")
AddAutoloadDir("D:\A encoder\simps_s02_60FPSv3\simps_s02_60FPSv3_plugin\SMDegrain v3.5.3d")
SetMemoryMax(8192*3)
SetCacheMode(0)

chemin = "D:\A encoder\simpsons s02_enc\Simpsons_S02E01.d2v"

d2vsource(chemin, idct=5)


RequestLinear(rlim=50, clim=100, elim=25)
# RequestLinear(rlim=60, clim=100, elim=40)

RoboCrop(LeftAdd=2, TopAdd=0, RightAdd=0, BotAdd=0)
Convertbits(16)
QTGMCp( Preset="Very Slow", SourceMatch=3, Lossless=2, Tr2=3, Nnsize=3, NNeurons=3, EdiMode="BWDIF+NNEDI3").Selecteven()
TDecimate(Mode=1)

### Color Conversion ###
fmtc_bitdepth(bits=32, fulls=False, fulld=True)
fmtc_resample (css="444")
fmtc_matrix (mats="601", col_fam="RGB")
fmtc_transfer (transs="1886", transd="linear")
#~ fmtc_primaries (prims="470bg", primd="709")
fmtc_primaries (prims="601-525", primd="709")
fmtc_transfer (transs="linear", transd="1886")
fmtc_matrix (mat="709", col_fam="YUV")
fmtc_resample (css="420")
fmtc_bitdepth(bits=16, dmode=1, fulls=True, fulld=False)

### Adjust Color ###
MergeChroma(aWarpSharp2(Depth=14))


### Gibbs Noise Block ###
a=last
ex_boxblur(0,1,mode="mean")
turnright()
ex_median("verticalS")
turnleft()
ex_median("IQM")

# Valley Mask
edge=FlatMask(1, scale=10).removegrain(2)
Mask=ex_makediff(Edge.MT_Inflate().ex_Expand(4).ex_gaussianblur(6,UV=1),Edge.MT_Inflate(),dif=false).Blur(1.0) #.ex_gaussianblur(6,UV=2)

a.MT_Merge(a.ex_median("EMF").ex_median("EMF").ex_Minblur(1).ex_median("STWM"),Mask,Luma=True)

# ex_adddiff(mask.converttoyuv420(), dif=False)


### Derainbow ###

masklisse = ex_median("IQM").blur(0.25).Removegrain(17).Flatmask(lo=4,MSR=80,scale=8).ex_inflate(2).Removegrain(7).Removegrain(17)    #.maa2().ex_median("STWM") #.ex_expand(1,"disk").ex_inpand(1,"disk")#.mt_binarize(threshold=25).ex_smooth(2,sharp=true)
masklisse = masklisse.ex_Expand(2, "disk").ex_median("STWM")
MT_Merge(last , ChubbyRain3().DFMDeRainbow(maskthresh=10, mc=False, use_Flux5framesT=False).ex_median("STWM"),masklisse,Luma=True)
#~ return ex_adddiff(masklisse.converttoyuv420(), dif=False)

### Overall Temporal Denoise ###
ConvertBits(bits=8)
#~ TemporalDegrain2(degrainTR=3, grainLevel=2, postFFT=5, postSigma=0.75, fftthreads=8,degrainPlane=0)
TemporalDegrain2(degrainTR=4, grainLevel=3, postFFT=5, postSigma=0.75, fftthreads=8,degrainPlane=0)
ConvertBits(bits=16)

### Resize ###
Crop(2,2,-2,-2)
deep_resize(960, 720,flat="nnedi3",grain=0,qual=1)

aWarpSharp2(type=1,Depth=4)
Sharpen(0.2)

### Darken-Thin Lines ###

#~ F=last
#~ S=F.FastLineDarkenMod(Strength=24,Prot=6).aWarpSharp2(Blur=4,Type=1,Depth=9,Chroma=2)
#~ D=ex_MakeDiff(S,F)
#~ ex_adddiff(Last,D,Dif=True,UV=2)

F=last
S=F.FastLineDarkenMod4(Strength=72,Prot=36,thinning=36).halobuster(thr=2, elast=2).aWarpSharp2(Blur=4,Type=1,Depth=4,Chroma=1)
D=ex_MakeDiff(S,F,UV=1)
ex_adddiff(Last,D,Dif=True,UV=2)

# F=last
# S=F.FastLineDarkenMod4(Strength=72,Prot=36,thinning=36).halobuster(thr=2, elast=2).aWarpSharp2(Blur=4,Type=1,Depth=4,Chroma=1).DRK(10)
# D=ex_MakeDiff(S,F,UV=1)
# ex_adddiff(Last,D,Dif=True,UV=2)

### Resize ###
deep_resize(1440, 1080,flat="nnedi3",grain=0,qual=1)

# aWarpSharp2(Blur=16,Type=1,Depth=3,Chroma=4)

# LSFplus(preset="slow", strength=50, smode=5, edgemode=0, preblur="""ex_median("IQM")""")
LSFplus(preset="slow", strength=150, smode=5, edgemode=0, preblur="""ex_median("IQM")""")

### Deband ###
# GradFun3plus(ThR=0.45,SMode=2) # 0.45 ep 1 à 6
# Merge(GrainFactory3mod(),0.225) # 0.225 ep 1 à 6

GradFun3plus(ThR=0.55,SMode=2) # 0.45 ep 1 à 6
Merge(GrainFactory3mod(),0.3) # 0.225 ep 1 à 6



# trim(9000,12000)
# trim(9000,9200)
trim(9000,10000)
# Prefetch(12,8) # 1.491
# Prefetch(4,8) # 1.096
# Prefetch(8,8) # 1.254
# Prefetch(12,12) # 1.557
Prefetch(2,8) # 


Edit 3:

I have now split the code in 2 pass, just after the Gibbs noise block.

I can run 5 encode simultaneously without problems. At least in some tests. I will let it run over night to see what happens.

Yesterday I ran 3 encode at same time in one script and I had no problem. ¯\_(ツ)_/¯

I someone have input as to why it could happen I'm open.

Also I know the script seems intense, and it is. I'm open about it. But they get pretty clean. At least cleaner than Disney+