FILM9 - Restoration software for old films (8mm, Super8, 16mm) and video

[lollo]

It was agreed that I would not use Neat for these test, and I did not use it.

Sorry, my mistake. However, whatever you used, the denoise was too aggressive.

And I'm not nostalgic for the quality of those old VHS. But I respect your choice which is worth mine.

I am biased at it, so my judgement can be arguable, but I still think AviSynth produced better result. Here 2 frames extracted from your side by side comparison.


[Attachment 65081 - Click to enlarge]


[Attachment 65082 - Click to enlarge]

The script I used uses only few lines and is very simple:

"levels and colors correction before filtering"
StabMod()
TemporalDegrain2()
LSFmod(ss_x=1.5, ss_y=1.5, secure=false, Spwr=2, SdmpHi=0, soothe=false, keep=25, edgemaskHQ=true)
"levels and colors correction after filtering"

In any case, thanks for your effort, a comparison of different methodologies is always constructive! I leave to the other users further comments

[lordsmurf]

I am biased at it

Not really. Bias is about ulterior motive, or undue preference. You have data that clearly shows the superiority of Avisynth, and the overaggressive damage by the non-Avisynth method. It's not bias, just good science.

Thanks for your samples. It always help to show what the rest of us know.

[Gelinox]

https://vimeo.com/715129666

Really so bad?
People will make up their own minds if they have an interest in it.
It does not matter, I did not expect any other reaction.

In fact, who can do more, can do less.
I know my preference is for cleaner visuals with less noise. This is my great fault (or my quality?).
I'm sure that if lollo had used the same program as me, he would have found the setting that would have suited him in a few clicks instead of the three lines of his script (by the way, are there any lines missing from this script? )

For me, the demo is done, you can find other programs than Avisynth that work just as well.
This does not mean that these dlls should be thrown in the trash, but to remain open to other technical possibilities that are easier to implement for the user.

Good continuation to all!

[lollo]

I'm sure that if lollo had used the same program as me, he would have found the setting that would have suited him in a few clicks instead of the three lines of his script

As I said earlier: "AviSynth has everything that is necessary for "problem fix/restoration needs" that I met". And I think it showed its power in the comparison

by the way, are there any lines missing from this script?

The full script is here, nothing "functional" was hidden in the shorted 5 lines I posted before:

"levels and colors correction before filtering"
StabMod()
TemporalDegrain2()
LSFmod(ss_x=1.5, ss_y=1.5, secure=false, Spwr=2, SdmpHi=0, soothe=false, keep=25, edgemaskHQ=true)
"levels and colors correction after filtering"

video_org=AviSource("t2a_sigla_1a_dtbc_nr_on_best_on_copie_off_d3r_on.avi")

video_org_trim=video_org.trim(988,3191)

video_org_trim_crop=video_org_trim.crop(8,12,-8,-12)

noise_baseclip=video_org_trim_crop.ColorYUV(off_u=5, off_v=-5).Tweak(sat=1.3, cont=1.08, coring=false, dither=true)\
.Levels(17, 1.05, 255, 16, 235, coring=false, dither=true)

stabilized=noise_baseclip.convertToYV16().StabMod()

denoised=stabilized.TemporalDegrain2()

sharpened=denoised.convertToYV12().LSFmod(ss_x=1.5, ss_y=1.5, secure=false, Spwr=2, SdmpHi=0, soothe=false, keep=25, edgemaskHQ=true).convertToYUY2()

corrected=sharpened.Levels(5, 1.0, 255, 0, 255, coring=false, dither=true)

video_restored=corrected.addborders(8,12,8,12)

return(video_restored)

For me, the demo is done, you can find other programs than Avisynth that work just as well.

For this excersize, not really

but to remain open to other technical possibilities that are easier to implement for the user.

I absolutely agree!

Good continuation to all!

Good continuation to you as well!

[Alwyn]

I agree with you, Gelinox. 90% of the result for 10% of the effort.

[lollo]

10% of the effort

I know that you are not familiar with AviSynth, but understanding, writing and running that script does not require any more effort than install, understanding and running a new NLE from scratch

It is similar to the discussion we had here http://www.digitalfaq.com/forum/video-capture/12549-deinterlacing-video-sharing.html

[GillesH]

Hello lollo,

Just curious, please.
- with which Avisynth version do you work ?
- with your script TemporalDegrain2, and on a VHS PAL 720x576 standard, what is the processing time for 1mn of clip and with an uncompressed encoding ?
Of course, I imagine it also depends on the power of the CPU.
Thanks for these informations.

[lollo]

with which Avisynth version do you work ?

AviSynth 2.6 build:Mar 31 2015

with your script TemporalDegrain2, and on a VHS PAL 720x576 standard, what is the processing time for 1mn of clip and with an uncompressed encoding ?

On my laptop Asus ZenBook UX530UX-FY033R, Windows 10 Pro that script runs at ~3 frames per second to generate a HuffyYUV output through VirtualDub.

If feeding ffmpeg with the avs script to compress to h264, with

ffmpeg.exe -i t2a_sigla_1a_dtbc_nr_on_best_on_copie_off_d3r_on_v21y.avs -c:v libx264 -crf 17 -preset slow -aspect 4:3 -c:a aac -b:a 128k t2a_sigla_1a_dtbc_nr_on_best_on_copie_off_d3r_on_v21y.mp4

it runs at ~2.6 frames per second

[johnmeyer]

I've used Neat many times over the almost quarter century I've been restoring videos and film. I agree that it is infinitely easier to use than AVISynth, but despite that, I have never once preferred the result it produces when compared to what can be done with AVISynth. The reasons are shown in these comparisons, namely that when you set it to remove enough noise to get meaningful noise reduction you also see a lot of detail reduction. It has always seemed to act more like a spatial denoiser than a temporal denoiser, something that is a disadvantage when trying to remove all those small noise dots that are characteristic of VHS noise or film grain.

I should also point out that any comparison of noise reduction needs to include the moving images rather than just comparisons of single frames. One of the other issues with noise reduction is the artifacts it adds to the video which can only be seen when the actual video is viewed. In many cases these artifacts can be more distracting and troublesome than the noise which was reduced.

Finally, the settings used make a huge difference. Every person (including me) when they first start doing the work always gets giddy from seeing the noise go away and therefore keeps turning up the settings until no noise remains. They often don't even notice, at least initially, at how much violence has been done to the integrity of the original video. Remember, you are performing noise reduction, not noise elimination.

The best noise comparisons that I have seen are always done by one single person using each of the different tools being compared. That way you avoid introducing into the end results conflicting tastes and widely disparate levels of skill.

Thanks to everyone for these tests.

[GillesH]

OK lollo,
Thanks

[lollo]

Thanks to you johnmeyer for your useful post!

We should also add that AviSynth allow to repair some "damage" not fixable in a easy way by a NLE:

https://forum.doom9.org/showthread.php?p=1944708#post1944708

https://forum.videohelp.com/threads/403553-Restoring-VHS-Tapes-with-Damage-%28White-Ho...-Distortion%29

[johnmeyer]

We should also add that AviSynth allow to repair some "damage" not fixable in a easy way by a NLE

One of the clips that I provided in those discussions you linked to provides a perfect example of what I'm talking about. In those discussions we were trying to eliminate dirt spots from film transfers but, with the wrong settings or the wrong algorithm, the ball being thrown back and forth also gets removed.

[Alwyn]

I know that you are not familiar with AviSynth, but understanding, writing and running that script does not require any more effort than install, understanding and running a new NLE from scratch

I am familiar with AVISynth. I use it to correct VHS colour issues (thanks Jagabo!) and deinterlace my VHC captures (QTGMC). That is why I say it is easier to use Neat Video for noise reduction. If I have multiple clips in a movie (with titles, cropping, effects) that have varying noise levels, it is far easier just to apply NV to each clip than go through the rigmarole of doing multiple AVISynth file operations on multiple individual clips then bring them into my movie. 90% for 10%. And NV does not "overagressively damage" video any more than AVISynth would in the wrong hands.

[Gelinox]

I posted on Viméo an extract of clip which compares the restoration of lollo and mine.

This clip was only watched by two people and the average watch time was only 70% of the clip's length.
So, there are clearly people who give their opinions without having seen anything.

Very concretely, watch this clip and tell me what details have been deleted.

https://vimeo.com/715129666

[richardpl]

Many posters on this forum are simply already biased toward avisynth. So no point in changing already formed minds.

[Alwyn]

By using this site (sorry in French language), but you just have to put the url address:
https://telecharger-videos.com/vimeo

Thanks Gelinox, that works a treat.

[lollo]

Many posters on this forum are simply already biased toward avisynth. So no point in changing already formed minds.

True.

I am not all that expert about NLEs, but, expanding the discussion not only about some denoise and limited/light restoration as we did so far in this thread, until you show us how to solve with a NLE the hundreds of problems on the videos posted every months here and on doom9's and digitalfaq forums, there is no point in changing our minds

[lordsmurf]

Many posters on this forum are simply already biased toward avisynth. So no point in changing already formed minds.

What else is there?
I use Avisynth because it often works.

[Gelinox]

Sorry to post this message for the second time, but you have to speak loudly here if you want to be heard:

--------------------------------------------------------------------------------------------------------------------------------
I posted on Vimeo an extract of clip which compares the restoration of lollo and mine.

This clip was only watched by two people and the average watch time was only 70% of the clip's length.
So, there are clearly people who give their opinions without having seen anything.

>>> Very concretely, watch this clip and tell me what details have been deleted. <<<

https://vimeo.com/715129666
-------------------------------------------------------------------------------------------------------------------------------

Small clarification received yesterday, one of the people is my friend Gilles and the other.... it may have been me with my phone who looked at it

[lollo]

Your stats must be wrong, probably only refer to people logged with an account. At least one more person watched, downloaded and compared the videos, posting the results, me

And after watching for the nth time, i still find avisynth result much better (but I am sort of biased, we know )

[Gelinox]

And after watching for the nth time, i still find avisynth result much better (but I am sort of biased, we know )

Yes, there is no doubt.

Still no answer to my question...
I repeat for the xth time:

https://vimeo.com/715129666
>>> Very concretely, watch this clip and tell me what details have been deleted. <<<

[lollo]

I showed some detail loss in the pictures I posted (within the limits of mp4 compression first and jpg later). Your second clip is shortened, no major difference there in term of details, but do you seriously prefer your results to mine? You must be biased ad well

I will leave more comments to others, this time for real I promise!

[Gelinox]

I showed some detail loss in the pictures I posted (within the limits of mp4 compression first and jpg later). Your second clip is shortened, no major difference there in term of details, but do you seriously prefer your results to mine? You must be biased ad well

No, I don't prefer this clip extract to yours... because except for a more exposed image... it's very much the same.
Doing exactly the same thing is very difficult, and that's what I tried, but you're not the master of restoration.
I simply tried to show you that we can do the same job as you.
Again, if you had used my tools, you would have gotten what you like easily and quickly.
There is a part of interpretation, no rule of good taste takes precedence over another. It's a matter of personal choice.
You may want to keep the original aspect and the video noise, be the least aggressive as possible (like John), but you may also want to obtain a result that is close to the camera that filmed the animation. When your video cassette left the store, it did not have all these defects.
Your work is also an interpretation, not an exact science.
I repeat it is a choice, not a rule.

[lollo]

you're not the master of restoration.

With all due respect, this is becoming ridiculous. I never said I am the master of the restoration, I just try to improve a little bit my vhs captures.

Again, if you had used my tools, you would have gotten what you like easily and quickly.

I respect your work, but why on earth should I use your tool to obtain inferior results? Is your tool a cosmetic change to videofred's script? A customized NLE? Something else? I thought we were comparing results, not promoting your tool. I am not interested in that.

Even for a newbie it takes 5 minutes to write and run that basic/simple TemporalDegrain() + LSFmod() script in AviSynth, there is no magic there; except for people who do not want to spent that time and prefere a GUI for whatever dedicated software (ffmpeg is an example), giving up to have the full control of the processes and fine tuning. And for advanced user you have available all the power of the AviSynth filters to play with the parameters and obtain the best possible result. I aim to the best possible results, good "enough" is not good for me. Speed and frame per seconds is not my priority either.

Your work is also an interpretation, not an exact science.

True, but for many of my captures I also have a direct DVB-S dump of the same broadcasted TV program. So I can compare what the video should be (a dvb-s recording is a bit-by-bit copy of what is transmitted) and perfectly recognize and fix the defects introduced by the analog recording going from dvb-s set-up box to analog vcr to capture card.

And finally, thanks to you and your friend Gilles for this discussion, it is always constructive to have a comparison. See you next time!

[Gelinox]

you're not the master of restoration.

Google translate changed the meaning of what I wanted to express, but it doesn't matter, at least for me.
Sorry if this hurt you, it was neither my intention nor even the meaning of my remarks.

I respect your work, but why on earth should I use your tool to obtain inferior results? Is your tool a cosmetic change to videofred's script? A customized NLE? Something else? I thought we were comparing results, not promoting your tool. I am not interested in that.

We have absolutely nothing to sell, no mystery tool to promote.
"My tool" only referred to the NLE versus "your tool" which is Avisynth, nothing more.

We will remain on different opinions:
- You think that Avisynth is the only tool that can allow you to restore your videos.
- I think you are missing some interesting experiences.

I don't think it will be possible for very long, but I wish Avisynth a long life.

[ConsumerDV]

you're not the master of restoration.

With all due respect, this is becoming ridiculous. I never said I am the master of the restoration, I just try to improve a little bit my vhs captures.

I did not call myself "new expert" either. People assume or invent personal traits to confront personalities not opinions. This is common and boring.

Even for a newbie it takes 5 minutes to write and run that basic/simple TemporalDegrain() + LSFmod() script in AviSynth, there is no magic there; except for people who do not want to spent that time and prefere a GUI for whatever dedicated software (ffmpeg is an example), giving up to have the full control of the processes and fine tuning.

Long-time users know certain settings and can type in the numbers without previewing, this does not mean that it is more convenient. Do professional colorists use command-line tools? I think most of those who work on big projects instead of batch-converting old VHS recordings use WYSIWIG tools because they can dial exactly the look they want and they can see the changes right away.

[lollo]

I did not call myself "new expert" either.

That was a joke in the other thread. No personal attack. Your nickname was created by DB83 first time you arrived and used in several other posts

because they can dial exactly the look they want and they can see the changes right away.

And why do you think that this is not possible with AviSynth?

[ConsumerDV]

And why do you think that this is not possible with AviSynth?

Because Avisynth is a scripting tool. Loading an avs script into VirtualDub does not make Avisynth a WYSIWYG tool.

[Gelinox]

Waoooo, a WISIWIG app for Avisynth would be great !!! You should invent that!!!

[poisondeathray]

And why do you think that this is not possible with AviSynth?

Because Avisynth is a scripting tool. Loading an avs script into VirtualDub does not make Avisynth a WYSIWYG tool.

You can use avspmod. Sliders to dial in settings. Programmable sliders as well. It's essentially WYSIWYG. Changing a slider, causes instant refresh like a NLE.

What's missing is keyframable settings, interpolation curves