Why is Neat video the best video denoiser?

[Truthler]

But why didn't you chose frames where peaking assist are not or less present?

Jumped through the source to see whether the on-screen-display content was through out the whole clip (and it was) and then jumped back to a frame where the 'peaking assist' dots were good visible to be sure it was clear what I was refering to when I asked about the 'red dots'.

Hello Selur!

Why don't you selected the ISO 10000 very noisy part of the video instead of the cleaner ISO 320?

The other problem I don't know which frame was denoised in your picture. The number of the frame is important if you post pictures instead of video.

[Selur]

I did not denoise or filter the clip with the osd info.
Didn't see much difference throughout the C0112 clip, last screenshot I posted was frame 267 in case you are interessted.
Like the rest of the readers I'm still not clear what you meant with the ballpen comment.

[Truthler]

I did not denoise or filter the clip with the osd info.
Didn't see much difference throughout the C0112 clip, last screenshot I posted was frame 267 in case you are interessted.
Like the rest of the readers I'm still not clear what you meant with the ballpen comment.

OSD and tech info was necessary, because some user said, that they want to see the method of noise generation in the original video.

But I can turn off the peak colors.

[sophisticles]

I don't get this discussion at all. The OP posted what sounded like an advertisement for Neat Video, at some point Lord Smurf, who used to swear by Neat Video, joined in to say it sucks, the thread went on for pages, at some point the OP said he would post a video that needed denoising so that we could compare Neat Video to other methods, then he posts this trash that needs to be deleted not denoised and here we are.

[lordsmurf]

at some point Lord Smurf, who used to swear by Neat Video.

Never. It was always a machete when you wanted a scalpel. I did infrequently use it 20 years ago, when Avisynth was more problematic and less rewarding. But VirtualDub and TMPGEnc were the weapons choice back then, not Neat Video.

It does appear to have matured finally, in more recent versions. Perhaps I'll reevaluate when time permits. But I heavily doubt it will outdo the multiple filters available in Avisynth. It's still quite limiting overall. Perhaps in certain narrow uses, maybe, but I doubt even that.

then he posts this trash that needs to be deleted not denoised and here we are.

Pretty much, yep.

[Truthler]

The triumph of quality.
Denoise with NEAT VIDEO.

You can download it here: 2GB

https://sendgb.com/PUHYg7hk0Eb

[Truthler]

at some point Lord Smurf, who used to swear by Neat Video.

Never. It was always a machete when you wanted a scalpel. I did infrequently use it 20 years ago, when Avisynth was more problematic and less rewarding. But VirtualDub and TMPGEnc were the weapons choice back then, not Neat Video.

It does appear to have matured finally, in more recent versions. Perhaps I'll reevaluate when time permits. But I heavily doubt it will outdo the multiple filters available in Avisynth. It's still quite limiting overall. Perhaps in certain narrow uses, maybe, but I doubt even that.

then he posts this trash that needs to be deleted not denoised and here we are.

Pretty much, yep.

lordsmurf

I've found a video for you, which explain how to capture VHS (yes Yes with OBS) in high quality. At least it explains you the basics. Good learning!
https://youtu.be/Y7PoNzrECG4

[Selur]

Where is the information about the settings used in Neat Video so that this can be reproduces by others?

I really like the text recovery, but the general denoising isn't good at all.

[Truthler]

Where is the information about the settings used in Neat Video so that this can be reproduces by others?

I really like the text recovery, but the general denoising isn't good at all.

The main point was the clarity of text, because it is very unequivocal visually. Don't forget the quality of the handwritten text too.

I selected on the video 4 noise varians with the mouse on the most noisest parts.

Only that people can reproduce his, who purchased the latest version with full funcions (incl. 4K and 8K video support)
But it is off-topic.
What does mater is to reproduce these quality (or making a better quality denoise) with free softwares.

[_Al_]

Thrutler - man, you are erratic as hell, its pretty simple, post:
1. an original footage
2. and in the same post, fixed video by NeatVideo.
3. if someone tells you that makes no sense, post something again

[Selur]

The main point was the clarity of text,...

Okay, would have been nice to know that.

You also did not answer the question.

btw. Are you sure you used the same source as you shared with us since when I load both files and look ad them side by side the content does not quite match:
Frame_0

Frame_100

Frame_200

Frame_300

Frame_400

Frame_500

Frame_600


This seems strange since denoising a video should not move the content around.
So the question still stands:

Where is the information about the settings used in Neat Video so that this can be reproduces by others?

script I used for the screenshots:

# Imports
import vapoursynth as vs
# getting Vapoursynth core
core = vs.core
# Loading Plugins
core.std.LoadPlugin(path="I:/Hybrid/64bit/vsfilters/SourceFilter/LSmashSource/vslsmashsource.dll")
# source: 'E:\C0112.mkv'
# current color space: YUV420P8, bit depth: 8, resolution: 3840x2160, fps: 29.97, color matrix: 709, yuv luminance scale: limited, scanorder: progressive
# Loading E:\C0112.mkv using LWLibavSource
clip = core.lsmas.LWLibavSource(source="E:/C0112.mkv", format="YUV420P8", cache=0, prefer_hw=0)
# making sure input color matrix is set as 709
clip = core.resize.Bicubic(clip, matrix_in_s="709",range_s="limited")
# making sure frame rate is set to 29.970
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# Setting color range to TV (limited) range.
clip = core.std.SetFrameProp(clip=clip, prop="_ColorRange", intval=1)
# adjusting output color from: YUV420P8 to YUV420P10 for x265Model
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P10, range_s="limited")
# set output frame rate to 29.970fps
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# Output
clip.set_output()

Cu Selur

[Truthler]

Thrutler - man, you are erratic as hell, its pretty simple, post:
1. an original footage
2. and in the same post, fixed video by NeatVideo.
3. if someone tells you that makes no sense, post something again

DEAR AI, I still can't understand what are you talking about.

[_Al_]

exactly, you are kind of slow, Selur spends days with you and so far nothing

[Truthler]

The main point was the clarity of text,...

Okay, would have been nice to know that.

You also did not answer the question.

btw. Are you sure you used the same source as you shared with us since when I load both files and look ad them side by side the content does not quite match:
Frame_0

Frame_100

Frame_200

Frame_300

Frame_400

Frame_500

Frame_600


This seems strange since denoising a video should not move the content around.
So the question still stands:

Where is the information about the settings used in Neat Video so that this can be reproduces by others?

script I used for the screenshots:

# Imports
import vapoursynth as vs
# getting Vapoursynth core
core = vs.core
# Loading Plugins
core.std.LoadPlugin(path="I:/Hybrid/64bit/vsfilters/SourceFilter/LSmashSource/vslsmashsource.dll")
# source: 'E:\C0112.mkv'
# current color space: YUV420P8, bit depth: 8, resolution: 3840x2160, fps: 29.97, color matrix: 709, yuv luminance scale: limited, scanorder: progressive
# Loading E:\C0112.mkv using LWLibavSource
clip = core.lsmas.LWLibavSource(source="E:/C0112.mkv", format="YUV420P8", cache=0, prefer_hw=0)
# making sure input color matrix is set as 709
clip = core.resize.Bicubic(clip, matrix_in_s="709",range_s="limited")
# making sure frame rate is set to 29.970
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# Setting color range to TV (limited) range.
clip = core.std.SetFrameProp(clip=clip, prop="_ColorRange", intval=1)
# adjusting output color from: YUV420P8 to YUV420P10 for x265Model
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P10, range_s="limited")
# set output frame rate to 29.970fps
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# Output
clip.set_output()

Cu Selur

Why should be an AVC mp4 contain exactly the same frame numbers when you transform it into Ut video avi file with voukoder in A. Premiere Pro?

[DB83]

There are a few things that I fail to understand about this discussion.


Firstly, I can not download either the 'before' or 'after' since I can not play 4K video smoothly. But then there does not appear to be any motion, if I read correctly, so 'video' is a bit of a mis-nomer here. But I assume that both, subject to the above remarks, are the same in length and 4K.


But this is now where I am confused.


1. When you shoot in good light you can set the ISO much lower than shooting in low light where you, theoretically, need the higher ISO to limit the noise. But here the ISO is being maxed to create noise ???? But if the created noise is 'artificial' then how can it be a fair test ?


2. I previously read that when dealing with noisy sources you need more bit rate than the original yet here the result is apparently less than 50% of that.


So maybe some knowledgeable, not necc. the OP, can enlighten me here.

[Selur]

Why should be an AVC mp4 contain exactly the same frame numbers when you transform it into Ut video avi file with voukoder in A. Premiere Pro?

What avc mp4 are you speaking of?
C0112.mkv is the file you shared as source, it UT lossless.
C0112_1.avi is the file you shred as filtered output.
Those two files both come form you. Both have the same frame count, but the content isn't the same.

the

# adjusting output color from: YUV420P8 to YUV420P10 for x265Model
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P10, range_s="limited")

at the end only changes the color space for preview


Also I made a mistake and posted the wrong script, should have been:
(my mistake)

# Imports
import vapoursynth as vs
# getting Vapoursynth core
core = vs.core
# Loading Plugins
core.std.LoadPlugin(path="I:/Hybrid/64bit/vsfilters/SourceFilter/LSmashSource/vslsmashsource.dll")
# source: 'E:\C0112.mkv'
# current color space: YUV420P8, bit depth: 8, resolution: 3840x2160, fps: 29.97, color matrix: 709, yuv luminance scale: limited, scanorder: progressive
# Loading E:\C0112.mkv using LWLibavSource
clip = core.lsmas.LWLibavSource(source="E:/C0112.mkv", format="YUV420P8", cache=0, prefer_hw=0)
# making sure input color matrix is set as 709
clip = core.resize.Bicubic(clip, matrix_in_s="709",range_s="limited")
# making sure frame rate is set to 29.970
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)
# Setting color range to TV (limited) range.
clip = core.std.SetFrameProp(clip=clip, prop="_ColorRange", intval=1)
# adjusting output color from: YUV420P8 to YUV420P8
clip = core.resize.Bicubic(clip=clip, format=vs.YUV420P8, range_s="limited")
# set output frame rate to 29.970fps
clip = core.std.AssumeFPS(clip=clip, fpsnum=30000, fpsden=1001)

filtered = core.lsmas.LWLibavSource(source="E:/C0112_1.avi", format="YUV420P8", cache=0, prefer_hw=0)
clip = core.std.StackHorizontal([clip, filtered])

clip.set_output()

---------

To make it clear: the content (view port, angle) of the source file and the filtered file does not match.



Cu Selur

[_Al_]

C0112.mkv is the file you shared as source, it UT lossless.
C0112_1.avi is the file you shred as filtered output.
Those two files both come form you. Both have the same frame count, but the content isn't the same.

Truthler - this is what I mean by saying "erratic" earlier. Rather post original and fixed video in the same post.

[poisondeathray]

Why should be an AVC mp4 contain exactly the same frame numbers when you transform it into Ut video avi file with voukoder in A. Premiere Pro?

What avc mp4 are you speaking of?
C0112.mkv is the file you shared as source, it UT lossless.
C0112_1.avi is the file you shred as filtered output.
Those two files both come form you. Both have the same frame count, but the content isn't the same.

C0112.mkv is AVC, not UT

It's not the same frame count - 601 frames vs. 588 frames

C0112_1.avi has jumps, looks like dropped frames .

It looks like the NV output was sharpened thus it's "noiser" - you can do the same thing apply sharpening to enhance text and make it more noisy too

[Selur]

@Truthler: Sorry, I was wrong about the two files.
But still I'm confused since it does not seem like you used the AVC as source, but something else.
-> if you did use something else than the C0112.mkv you shared with us than the whole test is flawed.

Cu Selur

Ps.: going to bed now.

[Truthler]

@Truthler: Sorry, I was wrong about the two files.
But still I'm confused since it does not seem like you used the AVC as source, but something else.

The FDR-AX700 uses AVC 100mbit mp4 format. I cut out the off-topic parts of the video with Avidemux, and I concentrated only on the part which contaioned the calendar, it was saved in MKV and I posted it on this forum with sendGB. The very same MKV AVC file was used to create the denoised video in Premiere with the help of Voukoder plugin. I saved it in UT video avi format. After I finished I post that denoised video too.

[poisondeathray]

You should check why you are dropping frames in premiere / voukoder

[DB83]

@Selur


This confusion is enough to send anyone to sleep.


@Truthler


Since these are your files then you can explain my concerns about ISO and bitrate (unless we are dealing with different 'video' since that is pretty unclear right now) and when your answers are published others can comment.

[Truthler]

@Selur


This confusion is enough to send anyone to sleep.


@Truthler


Since these are your files then you can explain my concerns about ISO and bitrate (unless we are dealing with different 'video' since that is pretty unclear right now) and when your answers are published others can comment.

There was an other video where manual IRIS ISO GAIN and other live camera data were present. That video was recorded in lossless ut video via Camlink 4K. I often use read peak for focus check and white overxplosure indicator in the camera. I forget to turn off the red focus peak indicator on that video, that's why Selur noticed it immediately in his comment.

The production method of the other so-called calendar video was explained clearly in my above coments.

[Truthler]

You should check why you are dropping frames in premiere / voukoder

I used max video memory in the HW acceleration settings of the Neat video, however during the denoising process I opened webpages in three different browser, many of them had YT 4K content, and news magaines etc.... Maybe the video memory caused that. Other: Something software problems with Premiere VOUKODER or, with Avidemux or I don't know.

It can have many causes.

[DB83]

Ok. I can understand that a different codec can have different bitrate requirements. Yet you originally put great emphasis on the original being lossless. That may not be a camera original but still...


But I still read zilch about the ISO issue.


Later today you have promised another 'video'. But if this is shot late in the day when people are not around so there is no actual movement how will this be any better than pointing a camera at an inanimate object. And if it is shot in daylight when noise should not be a natural issue what are we to think ?

[sophisticles]

what are we to think ?

That this is a complete waste of time?

This guy is shooting in a way that no one in his right mind ever would.

His video doesn't need a denoiser, it needs a basic class in how to use the proper camera settings and proper lighting, if he is using a two thousand dollar camera and this is what he is producing.

[sophisticles]

You should check why you are dropping frames in premiere / voukoder

I used max video memory in the HW acceleration settings of the Neat video, however during the denoising process I opened webpages in three different browser, many of them had YT 4K content, and news magaines etc.... Maybe the video memory caused that. Other: Something software problems with Premiere VOUKODER or, with Avidemux or I don't know.

It can have many causes.

No, it can't.

The cause is that you are trying to process video using a hardware accelerated filter and you simultaneously decided to open up multiple browser windows with 4k videos playing.

I don't know what kind pf pc you have but this is amateur hour, most systems will not be able to handle such a workload without dropping frames.

[lordsmurf]

I've found a video for you, which explain how to capture VHS (yes Yes with OBS) in high quality.

That video shows yet another cheapskate Youtuber low-end method. Deinterlaced, wiggly (no TBC), compressed, etc. You're essentially putting video in a blender, taking a poop on it, then trying to watch it. It's so far removed from the source, so non-quality, that it insults your eyes and ears to watch/hear. I'd rather just view something else, something not butchered.

The main point was the clarity of text,
...
What does mater is to reproduce these quality (or making a better quality denoise)

Those are completely separate activities. It's not an all-in-one task. That leads me to believe that you really don't know what you're doing, and are just moving sliders around the GUI without understanding restoration techniques.

1. When you shoot in good light you can set the ISO much lower than shooting in low light where you, theoretically, need the higher ISO to limit the noise. But here the ISO is being maxed to create noise ???? But if the created noise is 'artificial' then how can it be a fair test ?

That's actually a fair lab test for some things, when removing variables, just not whatever he's doing here. From a restoration POV, consumers often use(d) crappy cameras that under/overexpose. But that really has nothing to do with simple denoise, nor even half of the equation to a complex denoise. Just smacking it with denoise isn't restoring the clip. And restoring seems to be the unspoken goal, as there'd be zero reason to BOTH upscale+NR (and yet, forgetting to try to color correct). NR is also more than singular, and (for example) you must distinguish grain and chroma noise.

I previously read that when dealing with noisy sources you need more bit rate

Yes. H.264 just blurs the footage otherwise (MPEG and DV barfs out blocks). So trying to "sharpen" is ridiculous. Start with the least amount of compression possible. Otherwise the test is somewhat pointless. Sure, there are real-world scenarios where the source was dumped, and you're stuck with digitized butchery, but the possibilities are limited.

looks like dropped frames .

I'm shocked! /s

This guy is shooting in a way that no one in his right mind ever would.

That's not true. This is what it looks like when somebody sets the camera down, and forgets to turn it off.

this is amateur hour, most systems will not be able to handle such a workload without dropping frames.

That's very much the case here. Most serious video folks have dedicated systems to video. They don't try to use a single computer to appease their ADD-addled mind. Multitasking is essentially doing several things halfassed. Monotasking is doing one thing well. That's always been an easy tell for amateurish video technique, for decades now.

[Selur]

Trying to sum the whole thing up, before this turns in a bashing contest:

Assuming that Truthler used the same file as source that he shared with us is:
a. Truthlers setup is flawed (drops frames, doesn't follow the frame order even on the first frame).
b. At least I am not able to produce the text enhancement that Neat Video is capable on that source.
c. Sadly, Truthler didn't share a lossless source of the selected scenario like he said he would and the clip he shared wasn't really versatile.
d. Sadly, Truther didn't share his Neat Video settings so validating what he did, by other Neat Video users, is not possible.

So my conclusions so far are:
a. in case I want to restore some text in a video Neat Video might be worth looking at.
b. at least the denoising performance I can see on Neat Videos site and in Truthler filtered output does not impress me.
c. I shouldn't have started playing around with the uploaded file.


Cu Selur

[DMS]

Feel free to use the last capture I attached in this thread https://forum.videohelp.com/threads/401454-Best-settings-using-Virtual-Dub-to-correct-...rk-VHS-capture
I am keen to see how well it will clean it up.