Why is Neat video the best video denoiser?

[Selur]

No way that amateur Avisynth based denoiser can compete with it.... And I will prove it.

I'm happy if you are right and Neat Video will be clearly unbeatable, atm. I suspect Avisynth/Vapoursynth will provide the same or even better results, but requires more knowledge and work.
So I suspect that "Neat Video" has it's merrit, but I doubt it's something others can't compete with.
-> Looking forward to your lossless before- and after-sample with a quality Avisynth/Vapoursynth can't archive.

[_Al_]

http://macilatthefront.blogspot.com/2020/01/two-different-solutions-for-denoising.html
SMDegrain is worse, they guy had only a Neat video 5 trial version, but even the Trial proved to be better than the above mentioned free solutions...

But in that example he used NeatVideo5, right, but already processed with QTGMC, which you cannot do in NLE and it is avisynth/vapoursynth filter!

again what pdr says, when you come up with a solution in avisynth/vapoursynth, it is a chain of filters,
or when there is a question on this forum, how to fix a video, and it gets answered by a guy with tremendous knowledge and logic approach, it is hardly only one line but mostly a solution with certain logic and workflow. That logic and workflow might be totally different in other app (hidden behind sliders in UI as pdr says)

[poisondeathray]

http://macilatthefront.blogspot.com/2020/01/two-different-solutions-for-denoising.html
SMDegrain is worse, they guy had only a Neat video 5 trial version, but even the Trial proved to be better than the above mentioned free solutions...

But in that example he used NeatVideo5, right, but already processed with QTGMC, which you cannot do in NLE and it is avisynth/vapoursynth filter!

again what pdr says, when you come up with a solution in avisynth/vapoursynth, it is a chain of filters,
or when there is a question on this forum, how to fix a video, and it gets answered by a guy with tremendous knowledge and logic approach, it is hardly only one line but mostly a solution with certain logic and workflow. That logic and workflow might be totally different in other app (hidden behind sliders in UI as pdr says)

Valid point, and QTMGC denoises too. If you used the NLE deinterlacer , you'd have other kinds of noise, aliasing (aliasing is a type of unwanted signal - or noise - that NV doesn't handle as well)


So let's rephrase it - given a QTGMC output used as input - which denoiser does better on that source ? Just testing denoiser here

In that test, on a single screenshot NV looks cleaner... but "cleaner" isn't always better, it usually means loss of signal too. For example, it changes the eye color - the chroma denoise settings were probably too high. (Personally I would have left a bit of noise with NV , the overall result looks a bit overdenoised). But based on those single shots, I'd still say NV did better - but you don't know if he used ideal settings or scripts . It's difficult to say - because he didn't use ideal NV settings either ...and it's his own bloody eyes...! On his website it looks a bit like heterochromia, but either way the color has changed

Many people know this already, but it's important to post source and outputs. It's important to show more than 1 frame, because a denoiser might perform well on single frame, but flicker in motion, or do poorly on another. (I know neat video wouldn't do that on this type of source, but I'm just saying "in general" for good practices for thorough and transparent testing guidelines). Moreover, people with current version of NV could demonstrate up to date versions - that test was from >1 year ago. And you don't know if the author sampled correctly (maybe he could have used better frame or location), or if he used ideal settings, because surely the results could have been even better (E.g. don't use as heavy chroma denoise and kill the eye color). Same with the other denoisers.

[Alwyn]

Which NLE ? It probably accepts AVFS

Movie Edit Pro. It may well do; back in the day we used to frameserve AVI.VDRs into MEP from Vdub, but it's now completely 64bit and I can't get it working. I'll check out AVFS though.

Alwyn, it is time to try out modern Neat video too. Your experiences came from an old version maybe... 4-5 years old.

I'm right up to date with Neat Video. It's not the be all and end all; it does get tripped up occasionally, but from a cost-benefit point of view (where "cost" equals time and brainpower) it's great.

[poisondeathray]

Which NLE ? It probably accepts AVFS

Movie Edit Pro. It may well do; back in the day we used to frameserve AVI.VDRs into MEP from Vdub, but it's now completely 64bit and I can't get it working. I'll check out AVFS though.

The recent versions are bundled with vapoursynth. You can extract avfs from the "portable" versions . It works with .avs and .vpy scripts . The "fake" avi will appear in the c:\volumes subdirectory

avfs script.avs

[Truthler]

No way that amateur Avisynth based denoiser can compete with it.... And I will prove it.

I'm happy if you are right and Neat Video will be clearly unbeatable, atm. I suspect Avisynth/Vapoursynth will provide the same or even better results, but requires more knowledge and work.
So I suspect that "Neat Video" has it's merrit, but I doubt it's something others can't compete with.
-> Looking forward to your lossless before- and after-sample with a quality Avisynth/Vapoursynth can't archive.

Next week I will shot an uncompressed test video, I am searching big smooth metal surfaces (like cars at a parking place) and large smooth glass-metal objects (like modern office buildings) I will use high gain and fast shutter speed to generate enough noise.

[Selur]

Also make sure to use some lossless video format not some 'visual lossless' stuff. (source doesn't matter as long as everyone starts from the same, but the filtered clip should be real lossless)

[Truthler]

Also make sure to use some lossless video format not some 'visual lossless' stuff.

I will use UT video codec, since the lossless NVENC-HEVC needs very long GOP to achieve good compression, which makes the frame-to-frame step annoyingly slow in players. (I mean the step-back frames). I will keep the whole production chain in lossless way. Both source file and denoised file must be lossless.

[Selur]

Fine with me as long as it's lossless.

[Truthler]

Fine with me as long as it's lossless.

Selur, just some options from the NEAT video
Open the pictures in new page, just to see them in full resolution.








You can build a lot of surface profiles, select them individually and combine them

[Selur]

Atm. only thing that I find hand is the automated find/guess grain/noise option. Rest seems similar to what Avisynth/Vapoursynth (I usually use the later) offer.
We will see wether you can prove you claim:

No way that amateur Avisynth based denoiser can compete with it.... And I will prove it.

(I assume Vapoursynth is included here.)
As a side note: just because something isn't included inside some NLEs doesn't mean it's inferior or only for amateurs.
(Otherwise NLEs would never support new formats etc. since those are usually first used by developers, tinkers etc. )

Cu Selur

[lordsmurf]

...

I'm amused.

Removing standard shooting grain, or ISO noise, is a trivial trifle, especially with modern HD sensors. It that's all your doing, sure, Neat Video works well enough, no argument here from anybody. Most anything works well enough.

Filter some hellish VHS source, if you want to prove something.

[rrats]

Dumbest thing I've read on this thread was shooting at high iso just to denoise it later. High ISO already removes fine details. Extensive denoise settings destroy even more of them. What you're left with is a plastic looking video. Not to mention the processing power and time needed to fix badly recorded footage.

[Truthler]

...

I'm amused.

Removing standard shooting grain, or ISO noise, is a trivial trifle, especially with modern HD sensors. It that's all your doing, sure, Neat Video works well enough, no argument here from anybody. Most anything works well enough.

Filter some hellish VHS source, if you want to prove something.

Don't forget: I started digitalizing VHS 2 years before you.

[Truthler]

Dumbest thing I've read on this thread was shooting at high iso just to denoise it later. High ISO already removes fine details. Extensive denoise settings destroy even more of them. What you're left with is a plastic looking video. Not to mention the processing power and time needed to fix badly recorded footage.

Are you sure? I think it is able to restore details which were hidden due to noise. At first it sounds weird, but true. If you have tensor cores in your NVIDIA card, speed is not a big problem.

How Neat video do that? It uses some AI technology, It can calculte some details from the previous 5 frames....

[poisondeathray]

Anyone should feel free to include other common scenarios for testing too... It's good to look at various scenarios to see how a denoiser (or some filter) performs - strengths/weaknesses

[Truthler]

Anyone should feel free to include other common scenarios for testing too... It's good to look at various scenarios to see how a denoiser (or some filter) performs - strengths/weaknesses

Neat video uses some kind of AI technology, It can calculte some details from the previous 5 frames.... Where the information of the pixels were hidden due to the changing noise...it can detect the info , remove the noise and paste the info/pixel to its place.

I recorded a newspaper from a distance, and it restored the very small characters from the noise... clearly. Can the above mentioned technologies do the same? I think they are not able to do that.

[poisondeathray]

Neat video uses some kind of AI technology, It can calculte some details from the previous 5 frames.... Where the information of the pixels were hidden due to the changing noise...it can detect the info , remove the noise and paste the info/pixel to its place.

I recorded a newspaper from a distance, and it restored the very small characters from the noise... clearly. Can the above mentioned technologies do the same? I thionk they are not able to do that.

Not the current version, it's just motion compensation so far.

I think the next 6.x of Neat will have neural net / machine learning, because everyone is doing it. If everyone else is doing it, and you don't offer it, uniformed buyers will purchase the other product


Post your newspaper test

[Truthler]

Neat video uses some kind of AI technology, It can calculte some details from the previous 5 frames.... Where the information of the pixels were hidden due to the changing noise...it can detect the info , remove the noise and paste the info/pixel to its place.

I recorded a newspaper from a distance, and it restored the very small characters from the noise... clearly. Can the above mentioned technologies do the same? I thionk they are not able to do that.

Not the current version, it's just motion compensation so far.

I think the next 6.x of Neat will have neural net / machine learning, because everyone is doing it. If everyone else is doing it, and you don't offer it, uniformed buyers will purchase the other product


Post your newspaper test

Hi Poisondeathray.... I was wrong. AI is not the future in denoising. https://www.neatvideo.net/viewtopic.php?t=12582

[poisondeathray]

Hi Poisondeathray.... I was wrong. AI is not the future in denoising. https://www.neatvideo.net/viewtopic.php?t=12582

According to who ? The NV developer does not say that "AI" is not in the future...


Understand that very process of denoising already changes "details"

"lansing" comments are wrong - "AI" at it's core concept does not necessarily "create more details" - he misunderstands what a neural net actually does.

You can train a neural net to "create more details", but you can train a neural net to do other things too

Yes, current versions of "AI" denoising have various issues, but don't underestimate market driven forces and buyer driven behaviours

[rrats]

Dumbest thing I've read on this thread was shooting at high iso just to denoise it later. High ISO already removes fine details. Extensive denoise settings destroy even more of them. What you're left with is a plastic looking video. Not to mention the processing power and time needed to fix badly recorded footage.

Are you sure? I think it is able to restore details which were hidden due to noise. At first it sounds weird, but true. If you have tensor cores in your NVIDIA card, speed is not a big problem.

How Neat video do that? It uses some AI technology, It can calculte some details from the previous 5 frames....

Its not restoring details. It's making them up

[rrats]

Also again, for Gods sake, stop recording with high ISO settings

[_Al_]

There is nothing needed to produce a footage that needs to be fixed. Just browse history of this forum, where each week there is a video available for download.

So download those and fix them with NeatVideo.

[Truthler]

How would you call that? Totally unreadalble content became readable.

Pictures:

Look the picture in the blue frame and compare it with non-effected terriotries.

Open the full picture in a new tab/window

Picture 1


Picture: 2



I don't think that your freeware tools are capable to do that.

[Selur]

Got a file?

[Truthler]

Got a file?

How do you mean?

[Selur]

If you want to know whether this can be done with Vapoursynth&co you would need to share a file for others to look at it.

[Truthler]

If you want to know whether this can be done with Vapoursynth&co you would need to share a file for others to look at it.

This video is full with personal data. I will shot a different video in the coming days, as I promised.

I just wanted to show these pictures for the infidels, that their doubt is baseless and futile. It is the industry leading denoiser.

[Selur]

ok

[rrats]

Paid programs are just simple versions of complex free programs