Why is QTGMC so destructive and why do so many people still recommend it?

[sebastiaaan]

Hi there.

sorry for the rant but I just don't get it. I have a large collection of DVD's and spent a long time figuring out how to properly encode/deinterlace them. For some titles I just keep them as a Remux but the majority needs some enhancements (e.g. correcting AR, cropping, deinterlacing, etc.).
I kept hearing about QTGMC and for a long time I was actually one of the people who believed it to be the best deinterlacer out there. Until I started observing and questioning things.
Don't get me wrong, QTGMC is a great script and it's way more than just a deinterlacer. But for my purposes I don't think QTGMC is the best deinterlacer out there like so many people claim.
So to my first question: why is QTGMC so destructive?
I get it, the script has so many possibilities and it's a great denoiser. However for deinterlacing purposes I think NNEDi3 is still the best possible option out there.

I have a lot of DVD's which are very grainy (shot on film) or "suffer" from digital noise (shows shot on tape). I am a purist and I like to keep things as they are - and call me crazy but I actually like digital noise (well, it depends..) as it adds some texture to the content. Note that I'm mostly talking about content from the 80's to early 2000's so the noise (old sitcoms for example) look more like grain (luminance noise?) than the typical digital camera noise (chroma noise?).

I haven't found a way to make QTGMC look like the actual source or close to lossless. This is most likely because you can't turn off denoising completely. Trust me, I have spent countless of hours tweaking my settings with lossless presets, ezkeepgrain, grainrestore, noiserestore, sigma, etc.
At first glance it might seem everything is okay and looks good, but the reality is that QTGMC is "too smart" and there's no way to make it dumber.
For example no matter the settings you choose, the grain/noise will not stay the same.. The image looks sharper and cleaner, so does the grain/noise (it looks finer, like a DVD source grain compared to 1080p scan grain if you get me). But that is entirely artificial and shows that QTGMC isn't really retaining the grain, it adds artificial one.
And in my cases this has become very obvious and it looks very bad in some scenes. For example a wall looks grainy and when a person in front of it is walking or moving their head it kinda acts like a magnet to the grain making it appear very weird and it just looks off and inconsistent. I have noticed this across various sources as well.

I know QTGMC uses a mix of many different deinterlacers on a scene by scene basis such as NNEDI3, yadif, etc. so I dared to take the step to finally replace QTGMC with something else..
I always wondered why yadif in VLC was less destructive than QTGMC when obviously yadif is way worse (aliasing). That's because yadif is mostly just deinterlacing/interpolating and it does not really process the whole image.
I now completely switched to NNEDI3 and couldn't be happier to be honest. Yes, QTGMC might be doing a better job completely eliminating jaggies but so far I haven't noticed any problems or visual annoyances using NNEDI3, to me it's literally perfect and I wish I tried it sooner.

So my last question is: why do so many people still recommend using QTGMC when asked about a good deinterlacer? Am I just weird for liking grain or noise and trying to not mess with the picture as best as I can? I can't stand grainy (live-action) sources being heavily filtered, there's always some leftover grain/noise that will make the picture look completely inconsistent and horrible to watch, that's also why I can't stand low bitrate encodes (especially when viewing on modern screens).

There will probably be people who call me dumb and that I haven't fiddled with the settings enough to make it work (which I did, it's impossible imo). But why would I need to do that? Especially when people recommend QTGMC to noobs who most likely just use the standard presets because they don't want to spend time fine-tuning the settings.

I'm curious to hear other people's opinion about it as I haven't heard anything negative about QTGMC at all which confuses the heck out of me.
Thanks for reading and Merry Christmas to all

[Selur]

There is no: Best filter for all sources.
In your case a filter like YadifMod might be more suited.

I haven't heard anything negative about QTGMC at all which confuses the heck out of me.

Totally not my experience,...

[Sharc]

So my last question is: why do so many people still recommend using QTGMC when asked about a good deinterlacer?

It's mostly copy-paste-copy-paste and "you can't go wrong with it" recommendation rather than real own experience and tests I think. And there have been discussions for interlaced anime sources, for example.
That's no critism on QTGMC at all, but depending on the (noisy) source I prefer NOT to deinterlace at all but process and keep the footage interlaced - unless I need to apply vertical scaling or some odd vertical cropping for some reason. I found that not to deinterlace has been the best option - also in the sense of archiving - even though people sometimes question the quality of TV deinterlacers.

[sebastiaaan]

but depending on the (noisy) source I prefer NOT to deinterlace at all but process and keep the footage interlaced - unless I need to apply vertical scaling or some odd vertical cropping for some reason.

I'll always keep the original source somewhere and for some PAL content that's actually progressive (but encoded as interlaced) I tend to just keep them as is. But unfortunately this is a dream scenario, so many DVD's suffer from mastering errors and need proper cropping or aspect ratio fixes (ITU standard comes to mind). Also I mainly watch the content through Plex be it on my phone, tablet, TV or desktop web app. Not all of them support deinterlacing without transcoding and if they do it's mostly just yadif or whatever. So for true interlaced content I will do proper deinterlacing, this way I don't have to worry about anything. As I said, I'll always keep a copy of the original interlaced source but using so many different devices it's better for me if I just deinterlace beforehand

[Alwyn]

I just deinterlace beforehand

Given the topic of the thread, what do you use as your deinterlacer?

[Skiller]

I myself do not like QTGMC much – at default settings.

I spent... I don't know... years tweaking and refining it to find settings that make it close to transparent for my sources. And I pretty much suceeded.

You said if you make QTGMC keep the grain, the grain looks too fine and artificial. I had this problem as well. The key setting here is NoiseDeint="bob". And there are more which may not even seem like they are (as a side effect) related to noise (especially the TR and Rep settings).

The reason QTGMC is so "destructive" is due to the way the core works. It basically works like this: first it does a dumb bob-deinterlace (with or without more intelligent spatial interpolation such as NNEDI), then temporally blurs the differences in bob-shimmer away and uses motion compensation to detect where to do that and where not to avoid a motion blur effect. That was the original concept written by Didée and named TempGaussMC. It's a bob-shimmer remover. As you can imagine, the side effect of this approach is rather strong denoising. But luckily the script evolved dramatically over the years and allows for the side effects to be compensated – although not too many peoply seem to be offended by those.


Here is my honest recommendation of settings for keeping grain and making the deinterlacing quite transparent to my eyes.
Even if you are happy with NNEDI3 alone now, I would like to hear your thoughts on these settings.

QTGMC(TR0=2, TR1=2, TR2=1, Rep0=1, Rep1=0, Rep2=4,
\ DCT=5, ThSCD1=300, ThSCD2=110,
\ SourceMatch=3, Lossless=2, Sharpness=0.1, Sbb=0, MatchPreset="slow",
\ NoiseProcess=2, GrainRestore=0.0, NoiseRestore=0.4, NoisePreset="slow",
\ StabilizeNoise=false, NoiseTR=0, NoiseDeint="bob")

[johnmeyer]

As I have posted many times, deinterlacing always degrades the image, regardless of what tool you use. What is more, unless you are resizing, you don't need to do it, and your TV set will do a very good job.

(edit: there are a few other times you need to deinterlace, depending on the temporal/spatial way a particular filter works, but if you try your script on 10-20 seconds of fast moving footage, you'll quickly find out if you can get by without deinterlacing.)

[Selur]

As I have posted many times, deinterlacing always degrades the image, regardless of what tool you use.

which is basically true for most of the filtering one can do

What is more, unless you are resizing, you don't need to do it, and your TV set will do a very good job.

Got no TV set and I would have to see some data to believe the 'very good job' part on your avarage TV set.

[_Al_]

It's a flame question , so ...
..., what also happens, almost anyone who grabs that interlace video or DVD for re-encoding, (enhancing nowadays is almost a routine now), make it "bigger" for their TV etc., WILL screw it up anyway, encoding it as progressive or god knows how. Tiny percentage goes on this forum or knows how to do it properly, to deinterlace, so by spreading video already de-interlaced by QTGMC is not a bad idea at all.

For unique videos there should be original kept, that's without saying right at the same place as that double frame rate mp4 is. Never separate them, for it is not lost.

[lollo]

What is more, unless you are resizing, you don't need to do it, and your TV set will do a very good job.

Got no TV set and I would have to see some data to believe the 'very good job' part on your avarage TV set.

I agree; with my interlaced S-VHS capture, my Samsung UE46F8000SZ and my Sony A90J 4K produce inferior results compared to deinterlacing the video with QTGMC.
The same for any "media player" in my PC.

there are a few other times you need to deinterlace, depending on the temporal/spatial way a particular filter works

For temporal-spatial filter, if you want to keep the video interlaced, do a lossless deinterlace with Nnedi3(field=-2), the filtering and then interlace back at the end. Quite effective.

[johnmeyer]

Got no TV set and I would have to see some data to believe the 'very good job' part on your avarage TV set.

Do your own tests (borrow a TV set if you have to). Of course you have to know what to look for (do you actually know what poorly deinterlaced footage looks like??), but if you do, I doubt you'll be able to detect any artifacts.

[Selur]

Tested with a bravia tv set of my sisters against my 27" monitor a while back and at least to my eyes the manually deinterlaced content looked better,...
No shimmering, no too blured, no ghosting, color issues or similar, yes when using QTGMC you might loose some noise, but that does not bother me much. So if you say that TV sets nowadys are better than using Avisynth and Vapoursynth I believe you, but since I don't own a TV set I will still deinterlace (usually not with default settings).
(Back then I would have put the deinterlacing using by the TV set around the same as used by vpu hardware deinterlacers like NVIDIA and AMD use them.)

do you actually know what poorly deinterlaced footage looks like??

If I spot artifacts (shimmering, bluring, color issues, jagged edges,...) that should not be there (are nto present in the source) I would call it badly deinterlaced.
But if you can point me to a decent documentation I'm happy to educate me some more as I rarely have to deal with interlaced content to begin with and would by no means call myself a pro or advanced user when it comes to deinterlacing.


Cu Selur

[dellsam34]

I watched VHS digitized interlaced contents on numerous LED and OLED TV's I've owned and was not satisfied about the quality of the de-interlaced video they put out, There is a noticeable difference between the same software de-interlaced and TV de-interlaced contents, The same VHS contents played back directly from the VCR into TV via composite looked even worse to almost unwatchable, The build quality of TV seem to make a little difference but not to the extent to say some good TV's have better de-interlacers than say QTGMC.

The hardware de-interlacers that I'm aware of that can compete with software are the pro grade pitza box size scalers, High end videophile grade DVD and Blu-ray players/recorders with upscaling capabilities such as Oppo, Panasonic and some other brands, as well as some home theater stand alone scalers.

[Sharc]

As I have posted many times, deinterlacing always degrades the image, regardless of what tool you use. What is more, unless you are resizing, you don't need to do it, and your TV set will do a very good job.

(edit: there are a few other times you need to deinterlace, depending on the temporal/spatial way a particular filter works, but if you try your script on 10-20 seconds of fast moving footage, you'll quickly find out if you can get by without deinterlacing.)

As I understand spatial/temporal filtering issues can mostly be solved by filtering even and odd fields separately, then interleave back.

Another use case which requires bob-deinterlacing is framerate conversion from 25i to 30 telecine when I need to combine 25i video and 30p mobile phone clips in the same 30fps timeline or same file, for example.

[lollo]

As I understand spatial/temporal filtering issues can mostly be solved by filtering even and odd fields separately, then interleave back.

If you do:

separatefields()
temporalfilter or spatial/temporal filtering
weave()

you create a vertical misalignement.

Some links to old discussions:
https://forum.doom9.org/showthread.php?t=86394
https://forum.doom9.org/showthread.php?p=1617925#post1617925
https://forum.doom9.org/showthread.php?p=1618309#post1618309
http://forum.doom9.net/showthread.php?p=1921993#post1921993

[Sharc]

As I understand spatial/temporal filtering issues can mostly be solved by filtering even and odd fields separately, then interleave back.

If you do:

separatefields()
temporalfilter or spatial/temporal filtering
weave()

you create a vertical misalignement.

That's not what I do/said. I wrote that even and odd fields should be filtered individually and re-interleaved, then weaved, rather than just filtering separated top and bottom fields and weaving. That's not the same.

[lollo]

The same apply: working on odd and even fields in two separate videos is not good.

[Selur]

lollo: can you elabarate on that since at least now I understand the links you provided say that

SeparateFields, SelectEven.filter,SelectOdd.filter, Weave

alongside with
"JDL_UnfoldFieldsVertical, filter, JDL_FoldFieldsVertical"
and
"Bob, filter, SeparateFields, SelectEvery, Weave"
should be fine for spatial-temporal filters,...

[lollo]

- if we apply a "progressive filter" on an interlaced video (mode A) is not good:

interlaced frames video, 25 frames (50 fields) per second (25 frames i)
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A.........C.........E.........G (field 0) even lines
b..........d.........f.........h (field 1) odd lines
PAL50Ii --> field 0 and field 1 are even and odd lines of "camera" frames taken at different time [true interlaced video]

the filtering will be on the frame A+b, which is not the original frame because A and b comes from 2 different moments in time

- if we apply a filter on separated fields makes no sense

- if we apply a filter on separated fields + select even/odd (mode B)

SeparateFields() assuming TFF
[all video, progressive or interlaced, is frame-based; SeparateFields change it from frame-based to field-based]
["frame" count is doubled]
frame1 frame2 frame3 frame4 frame5 frame6 frame7 frame8
A.........b.........C.........d.........E........f .........G.........h

filtering (i.e. a denoiser) on A - C - E - G and on b - d - f - h will have half the frames to smooth the motion on, will have the ghosting as half as long and the temporal radius is halved by two.

There is a need for deinterlace if a filter may get the information of one field into the other field: sharpen, resize, antialiasing, motion compensation (that is everywhere) and if the fitler needs access to the true temporal resolution of interlaced video (for this last requirement, working on separated fields + select even/odd is ok)

- if we apply the progressive / spatial-temporal filter on the deinterlaced (eventually lossless) video is the ideal situation, and eventually we can interlace back at the end.

Some explanation better than mine :
https://forum.doom9.org/showthread.php?p=582734#post582734
https://forum.doom9.org/showthread.php?p=1618309#post1618309

[Selur]

Okay, so trying to summing things up for spatial-temporal filtering either:
a. bob->filter->reinterlace
Loose quality due to bobbing (which uses interpolation and often does not keep the original fields untouched), but assuming the interpolation is decent enough gain by better filtering.
b. separateFileds->filterEven&filterOdd->weave (+..)
Loose quality since the only half the temporal and spatial information is provided and filtering can't be as effective and might cause artifacts.
=> both methods are flawed and depending on the filters and scenes on or the other might be better.

Personally I always deinterlace since I filter and work (and filter) only progressive content, but if you want got interlaced content and want to output interlaced content depending on the content and filters one should compare the output of the methods.

Cu Selur

[Sharc]

... filtering (i.e. a denoiser) on A - C - E - G and on b - d - f - h will have half the frames to smooth the motion on, will have the ghosting as half as long and the temporal radius is halved by two.

This is perhaps the weak point of the even/odd grouping method, and may request a certain filter re-adjustment. The end result may still look better than blindly deinterlacing though - in my experience.

There is a need for deinterlace if a filter may get the information of one field into the other field: sharpen, resize, antialiasing, motion compensation (that is everywhere) and if the fitler needs access to the true temporal resolution of interlaced video (for this last requirement, working on separated fields + select even/odd is ok)

- if we apply the progressive / spatial-temporal filter on the deinterlaced (eventually lossless) video is the ideal situation, and eventually we can interlace back at the end.

Ideal only for an ideal deinterlacer (bobber), which is not realistic even when throwing QTGMC in, (referring to QTGMC's noise processing effects which have been addressed before, for example).

The 'right' method - apart from the beholder's eyes - will also depend on the source and on the particular filters, I think - once we exclude the obviously wrong principles.

[Sharc]

Okay, so trying to summing things up for spatial-temporal filtering either:
a. bob->filter->reinterlace
Loose quality due to bobbing (which uses interpolation and often does not keep the original fields untouched), but assuming the interpolation is decent enough gain by better filtering.
b. separateFileds->filterEven&filterOdd->weave (+..)
Loose quality since the only half the temporal and spatial information is provided and filtering can't be as effective and might cause artifacts.
=> both methods are flawed and depending on the filters and scenes on or the other might be better.

Personally I always deinterlace since I filter and work (and filter) only progressive content, but if you want got interlaced content and want to output interlaced content depending on the content and filters one should compare the output of the methods.

Cu Selur

Thanks for supporting both principles in Hybrid

[lollo]

Loose quality due to bobbing

We are filtering, so the bobbing issue (shift scanline) is not that relevant here, because spatially is on the next frame and temporally is averaged across the temporal radius and at the end we only keep 1 frame out of 2.
If we want to keep the output deinterlaced, than we use QTGMC in its full power (no lossless mode) for removing "bobbing effect", and apply any filtering after the deinterlacing.

and often does not keep the original fields untouched)

Nnedi3(fields=-2) is lossless and do keep the field that you keep, once you interlace back, untouched.
QTGMC in my experience is not that lossless, but QTGMC(lossless=1) in this case is ok.
You can verify with:

video_org=AviSource("<>")

deinterlaced=video_org.AssumeTFF().QTGMC(lossless=1)
interlaced=deinterlaced.AssumeTFF().SeparateFields().SelectEvery(4,0,3).Weave()

## check difference between original video and original video after QTGMC_lossless_1 and re-interlacing ###
difference_video_org_interlaced=Subtract(video_org, interlaced).Levels(65, 1, 255-64, 0, 255, coring=false)

but assuming the interpolation is decent enough gain by better filtering

Correct. Nnedi3 or QTGMC (which uses Nnedi3 internally) here.

both methods are flawed and depending on the filters and scenes on or the other might be better.

In my experience, and if I understand correctly the theory behind, "Deinterlacing method" is always better; experiment yourself

Personally I always deinterlace since I filter and work (and filter) only progressive content, but if you want got interlaced content and want to output interlaced content depending on the content and filters one should compare the output of the methods.

I agree, everyone should always experiment on his own material, rather then following what I or somebody else could say!

This is perhaps the weak point of the even/odd grouping method, and may request a certain filter re-adjustment. The end result may still look better than blindly deinterlacing though - in my experience.

I had a different experience. You are right about "blind" deinterlace: it is not appropriate, use a lossless "good" deinterlacer.

Ideal only for an ideal deinterlacer (bobber), which is not realistic even when throwing QTGMC in, (referring to QTGMC's noise processing effects which have been addressed before, for example).

See above, Nnedi3 and QTGMC lossless do wonders in this context.

The 'right' method - apart from the beholder's eyes - will also depend on the source and on the particular filters, I think - once we exclude the obviously wrong principles.

There is a theory behind the right method we are discusing, so the source of the video should not be relevant. However, as I said, it is always better to check and experiment from himself

[Selur]

QTGMC(lossless=1)

Why "lossless=1"? From how I understood it using lossless=2:

# Lossless=2 - after preparing an interpolated, de-shimmered clip, restore the original source fields into it and clean up any artefacts
    # This mode will not give a true lossless result because the resharpening and final temporal smooth are still to come, but it will add further detail
    # However, it can introduce minor combing. This setting is best used together with source-match (it's effectively the final source-match stage)

seems to be the better way to go.

[lollo]

They are equivalent for what we want to achieve

[Selur]

Nice, then the way I use it in Hybrid is okay.

[lollo]

I do not use it often, but Hybrid is a wonderful piece of software, so I am sure it is ok!

[Selur]

*gig* always working on it and fixing bugs, so it's probably not wonderful, but it's working towards it.

[poisondeathray]

I now completely switched to NNEDI3 and couldn't be happier to be honest. Yes, QTGMC might be doing a better job completely eliminating jaggies but so far I haven't noticed any problems or visual annoyances using NNEDI3, to me it's literally perfect and I wish I tried it sooner.

NNEDI3 is a single field deinterlacer only . When used alone for double rate deinterlacing, NNEDI3 does not compensate for the odd/even field ofset . You can end up with up/down/up/down motion (it might not be as noticable with a shaky camera, or lower quality/soft video). On adjacent frames, lines don't match up between objects - this one of the causes for NNEDI3 temporal aliasing and flickering artifacts. Similarly, NNEDI3 doesn't address the top/bottom frame border edges either - there is missing data on either the top or bottom frame edges on adjacent frames (because of even/odd scan lines) - QTGMC with border=true switch (not the default setting) compensates for that.


[Attachment 62572 - Click to enlarge]

This is a slowed down gif demo about 1/6 the speed (note in realtime 59.94p the artifacts will be less noticable to most people..., and on certain "organic" images with curves, not lines like buidings, fences etc.. - but the fact is there are expected artifacts because of how NNEDI3 works as an intrafield deinterlacer).

For QTGMC - The denoising and softening is by far the most common complaint. Another is the default sharpening strength is too high - when combined with the denoising at default settings, many sources look like "waterpaintings" or "plasticky." Other semi common QTGMC "negatives" - temporal artifacts. Sometimes the motion estimation is incorrect. e.g. ghosting issues. Another semi-common QTGMC complaint is the mvtools2 block discrepanies. Instead of being a single consistent image, some blocks or parts of the image might appear to move in slightly different directions (This can affect mvtools2 based denoisers as well).

Pros/cons to everything. Tradeoffs no matter what you do. The underlying problem is interlace in the first place - there is only so much you can do when 1/2 the lines are discarded. QTGMC is recommended for general use because in most cases it resembles the original 59.94p source the most (before the immediate source became interlaced) . A true progressive source shouldn't have that flicker or aliasing artifacts common to most deinterlacers.

[Selur]

The underlying problem is interlace in the first place

so true,...