Deinterlace challenge

[Sharc]

Is there a better method to double-rate deinterlace the attached (pathologic) source?
Some trials attached.

WARNING:
Watching this video may cause nausea or epileptic seizure to sensitive viewers.

[Alwyn]

My head's spinning!

[lollo]

My generic QTGMC setting shows less artifacts on the third region with horizontal only lines. Just a quick run, I did not experiment anything else...

QTGMC(preset="slow", matchpreset="slow", matchpreset2="slow", sourcematch=3, tr1=2, tr2=1, NoiseTR=2, sharpness=0.1)

comparison: comp.avi

[poisondeathray]

You should put an epileptic seizure warning

TDeint looks the least bad to me in that group

For this specific case you could "cheat" by using an lowpassed odd field luma mask + RIFE . You could overlay that on tdeint base with some temporalmedian applied. You'd expect the first few frames and last few to be messed up before it settles into the pattern per scene

[Sharc]

You should put an epileptic seizure warning

Yup, I missed it. I hope I won't be sued

TDeint looks the least bad to me in that group

For this specific case you could "cheat" by using an lowpassed odd field luma mask + RIFE . You could overlay that on tdeint base with some temporalmedian applied. You'd expect the first few frames and last few to be messed up before it settles into the pattern per scene

Amazing result with your approach tailored for this specific case
The source may be awkward and far from natural video but I think it reveals the basic weaknesses of deinterlacers, each one exhibiting its specific artefacts, actually "damaging" the source. One reason why I am usually hesitant to deinterlace for "archiving" purposes.
At the end I think we would need to view this on an interlaced display? Unfortunately today's progressive monitors do not (and most likely will not in future) offer a setting to behave like a field oriented interlaced display such as legacy CRT's.

(Btw it's also enlightening to stream the source to the TV to see how its deinterlacer behaves.)

[Alwyn]

it's also enlightening to stream the source to the TV to see how its deinterlacer behaves.

Mine is ugly, although that particular video isn't very realistic.

You should put an epileptic seizure warning

I hope I won't be sued

I think you should put a warning in your first post. That could be harmful to someone.

[Selur]

clip = havsfunc.QTGMC(Input=clip, Preset="Fast", InputType=0, TFF=False, TR2=0, SourceMatch=0, Lossless=2, opencl=True, Denoiser="KNLMeansCL", TR0=2, TR1=2, Rep0=2, Rep1=4, Rep2=8, DCT=5, Sbb=1, NoiseProcess=0, GrainRestore=0.0, NoiseRestore=0,  StabilizeNoise=False, Sharpness=0

should also work fine, main point is to disable sharpening, renoising and keep things lossless,...

[Sharc]

clip = havsfunc.QTGMC(Input=clip, Preset="Fast", InputType=0, TFF=False, TR2=0, SourceMatch=0, Lossless=2, opencl=True, Denoiser="KNLMeansCL", TR0=2, TR1=2, Rep0=2, Rep1=4, Rep2=8, DCT=5, Sbb=1, NoiseProcess=0, GrainRestore=0.0, NoiseRestore=0,  StabilizeNoise=False, Sharpness=0

should also work fine, main point is to disable sharpening, renoising and keep things lossless,...

The 1-pixel horizontal band on top is still flashing (line twitter black-white-black-white ......).
Anyway, not as a need a solution. I was just wondering whether I missed something trivial.

[Selur]

Which do you mean, looking at odd&even fields


isn't the top meant to flicker?

[jagabo]

The 1-pixel horizontal band on top is still flashing (line twitter black-white-black-white ......).

It's supposed to be flashing black and white. An interlaced analog TV shows one field at a time. Since one field is black and the other is white (in that section) the picture should flicker black and white.

You forgot a simple bob(0.0, 1.0) which is pretty much what one would see on a 59.94 Hz analog interlaced TV (lots of flicker). Careful though, some monitors and TVs will attempt to "fix" this video.

[Sharc]

isn't the top meant to flicker

The 1-pixel horizontal band on top is still flashing (line twitter black-white-black-white ......).

It's supposed to be flashing black and white. An interlaced analog TV shows one field at a time. Since one field is black and the other is white (in that section) the picture should flicker black and white.

You forgot a simple bob which is pretty much what one would see on an analog interlaced TV (lots of flicker). Careful though, some monitors and TVs will attempt to "fix" this video.

Makes sense, yes. (Btw a simple bob is included in my deint.mp4 as last case). I thought a legacy CRT may flicker less nervously because of its natural gradual brightness decay of the fluorescent screen. Maybe I am wrong. I don't have an old CRT anymore to verify. Anyway it was interesting to see how the various deinterlacers (and TVs) behave with this synthetic source.

[jagabo]

Btw a simple bob is included in my deint.mp4 as last case

Sorry, I missed that. Did you notice your deinterlaced video is 50p not 60p (from a 30p source)? So you are comparing a frame rate change as well as deinterlacing.

[jagabo]

Attached is a 60p version.

LWLibavVideoSource("source_IL.ts") 
AssumeBFF()

v0 = last.ChangeFPS(60000, 1001).Subtitle("interlaced_x2")
v1 = QTGMC().Subtitle("qtgmc")
v2 = BWDIF(field=2).Subtitle("bwdif")
v3 = yadifmod2(mode=1).Subtitle("yadifmod2")
v4 = tdeint(mode=1).Subtitle("tdeint")
v5 = nnedi3(field=2).Subtitle("nnedi3")
v6 = bob().Subtitle("bob")

v0+v1+v2+v3+v4+v5+v6

prefetch(8)

In truth, I don't think this is a good test video for deinterlacing. It's way too artificial.

[Sharc]

In truth, I don't think this is a good test video for deinterlacing. It's way too artificial.

Yes, that's almost exactly the script which I used. I just added AssumeFPS(25) at the beginning of the script for "PAL" framerate compliance - which makes the flicker more annoying.
I agree it's artifical-experimental and not a good test for practical deinterlacing. Yet one may find some of these 'deinterlacing defects' with anime sources or even 'natural' video.

[Bwaak]

An interlaced analog TV shows one field at a time.

Not really. It meshes the fields with one field being painted while another fades away, hence interlacing, not 60p. It is a progscan TV set that shows one field at a time in the basic line-doubled bob mode.

[jagabo]

An interlaced analog TV shows one field at a time.

Not really. It meshes the fields with one field being painted while another fades away

The previous field has mostly faded away long before the new field is drawn. If you look at very high speed photography of an analog TV screen you'll see that a scanline has mostly faded away even as the next scanline of the same field is being drawn. You can see an example here:

https://youtu.be/3BJU2drrtCM?t=207

If you watch an analog TV showing a video where one field is all black and the other is all white it flickers horribly.

[Bwaak]

@Jagabo, is it interlaced or progscan? Impressive footage.

[jagabo]

Early in that video (around the time I linked to) he is showing an inexpensive interlaced CRT TV. Later he shows a progressive LCD display where the situation is different.

[Bwaak]

Early in that video (around the time I linked to) he is showing an inexpensive interlaced CRT TV. Later he shows a progressive LCD display where the situation is different.

I was asking about the CRT. If it is interlaced, then those lines are pretty damn close.

I wish he "froze" one field and overlayed the next field onto it to show how the second field fills the gaps in the first one. But I get the point - the phosphors decay very fast, which is surprising to me. Thanks.

[Sharc]

...But I get the point - the phosphors decay very fast, which is surprising to me. Thanks.

The decay was much slower for old B&W TV's - some hundreds of microseconds as compared to some tens of microseconds of later color CRT TVs. It has to be seen in relation to the scanline duration. The basic effect is similar though.

[jagabo]

The scanlines are "close" because the electron beam isn't so highly focused that there's a gap between the lines. This makes the screen brighter and reduces the flicker (with a normal picture).

[Bwaak]

The scanlines are "close" because the electron beam isn't so highly focused that there's a gap between the lines. This makes the screen brighter and reduces the flicker (with a normal picture).

It depends. The beam could be more or less focused as needed.

A useless fun fact: with System A, as the size of TV screens grew, it was found that a wide beam makes the picture blurry and low-res, while a narrower beam leaves spaces between the lines. So a special technique was devised that used a narrow beam that wobbled up and down, producing oval-shaped spot. This allowed to preserve horizontal resolution and to close the gaps at the same time.