Search for alternative of some MSU Filters

[albesp77]

Dear All,
i want to know if there are some valid alternatives for using with AviSynth, possible with having good results, to those commercial MSU Filter:

http://compression.ru/video/frame_ra...x_en_frcn.html - Advanced Frame Rate Converter (AFRC)
http://compression.ru/video/old_film.../index_en.html - VirtualDub MSU Old Film Restoration
http://compression.ru/video/super_re...lution_en.html - MSU Super-Resolution Filter
http://compression.ru/video/super_re...cision_en.html - MSU Super-Precision Filter
http://www.compression.ru/video/deshaker/index_en.html - VirtualDub MSU Deshaker filter
http://www.compression.ru/video/grai.../index_en.html - Grain-Degrain Project
http://www.compression.ru/video/resa.../index_en.html - High quality image and video resampling

And another one for Deblocking, please.

Thanks to all peoples that reply to my question

Best Regards

[Hagbard23]

You can use the Filters (which are very good indeed) inside Avisynth via LoadVirtualDubPlugin ("filename", "filtername", preroll)

http://avisynth.org/mediawiki/Plugins

Keep in mind, that they do require a certain colorspace, which is most often NOT YV12

I do not know any substitutes to this filters, since they are very developed.

You can try those:
http://avisynth.org.ru/depan/depan.html [DESHAKE]
http://www.removegrain.de.tf/ [GRAIN REMOVAL]
http://avisynth.org.ru/docs/english/...s/addgrain.htm [GRAIN ADDITION]
http://forum.doom9.org/showthread.php?t=144271 [OLD FILM RESTORATION]
http://avisynth.org/mediawiki/FPS [FRAMERATE CONVERSION]
http://avisynth.org/mediawiki/LSFmod [LIMITED SHARPENING / SUPERSAMPLING / UPSAMPLING]

[Vitaliy Gorbatenko]

http://avisynth.org/mediawiki/External_filters

MSU Super-Precision Filter look's like frfun + GradFun2DBmod
High quality image and video resampling http://forum.doom9.org/showthread.php?t=147695
Deblocking - http://avisynth.org/mediawiki/Deblock_QED

[WILLIS]

I was actually going to post a similar thread about this topic. I was wondering how I go about using the MSU Smart Sharpen filter inside an avs script using the sharpen borders only preset...thx

[royia]

Any Super Resolution Filter (No upscaling just improvements of the Details and NR based on adjacent frames)?

[Hagbard23]

Any Super Resolution Filter (No upscaling just improvements of the Details and NR based on adjacent frames)?

I think those are worth a try...

http://avisynth.org/mediawiki/LSFmod

http://avisynth.org/mediawiki/LimitedSharpen

http://avisynth.org/mediawiki/SeeSaw

[royia]

They are not "SuperResolution" filters.
SuperResolution filters use Deconvolution process.

It's not Sharpening, it's Restoration.

[Didée]

Sweet dreams are made of this, ...

Read me! (once, twice, and if necessary, thrice.)


What is possible: >Such<, for example. (Basic filter kit, not optimized.)
Whether you can get there with MSU filters, I'm not so sure ... with Avisynth, you can anyway.

[Hagbard23]

Read me! (once, twice, and if necessary, thrice.)

Fully agree - thanks for that -cool- statement...

They are not "SuperResolution" filters.
SuperResolution filters use Deconvolution process.

It's not Sharpening, it's Restoration.

Go and spend some thousand dollars for senseless terms like "Super Resolution" - and you know what? You are still better off with the three free filters i mentioned (at least most of the time)

[royia]

Originally Posted by Hagbard23

Fully agree - thanks for that -cool- statement...

Go and spend some thousand dollars for senseless terms like "Super Resolution" - and you know what? You are still better off with the three free filters i mentioned (at least most of the time)

Sharpeners can only enhance the details which exists.
Super Resolution can restore details.
Those are 2 completely different operations.

I don't know about the Plug In's the OP mentioned for VirtualDub (Whether they are real and good Super Resolution Algorithms) I was just wondering if anyone developed one for AviSynth.

Didee, I read your post.
I don't agree with you.
Let's say I have one Blurred Frame.
Using Deconvolution I can restore Blurred details to some level.
This is Ill Poised problem, When of the toughest in Image Processing world yet it can be done to some extent.
And that's using one frame only and it's completely different from Sharpening.

As far as I know, SuperResolution just get more data for the "Deconvolution" process using adjacent frames.
Hence it should help restore "Blurred Details".

I don't know about the implementation in the Plug In the OP linked to. The above implementation might not use the Deconvolution process or doesn't do it well and hence you can't see improvements in the "Blurred Details".

I must say my knowledge is based on a Project I did about Deconvolution of Out of Focus Still Images.

Anyhow, if the user reports say they didn't notice "Blurred Details" improvement, I agree with your implication, no need to use it.

[Hagbard23]

So you mean, your "Super Resolution" "Deconvolution" Blah Blah, can make something visible, which doesn't exist anymore? ...I don't believe in that..but i will try it out, as soon as a good plugin goes my way...

[Didée]

A sharpen filter is a deconvolution filter. Even Avisynth's simple "sharpen(x)" does deconvolution.

No need to get lengthy again - here's my opinion on the matter of (blind) deconvolution.


Sure, there are plenty of "whitepapers". There even are whitepapers about constructing a perpetuum mobile.

[royia]

I see Didee.
I meant a Deconvolution after estimating the PSF.
I don't think that what's Avisynth does, Doesn't it?

If you make a good estimation of the PSF it works well in real life, not only on White Papers.
It can't do magic, Yet easily recover details no other Sharpening filter can.

Anyhow, I'll take your word on this.
I guess there aren't decent Super Resolution Filters for VirtualDUB / AviSynth.

Thanks.

[zee944]

Originally Posted by royia

It can't do magic, Yet easily recover details no other Sharpening filter can.

If you're so sure about the superresolution effect, could you show us a few good examples you've achieved with MSU Superresolution?

[royia]

Originally Posted by zee944

If you're so sure about the superresolution effect, could you show us a few good examples you've achieved with MSU Superresolution?

Did you read my messages?
Never used it.

I have no experience with Super Resolution.
My Latest Project on University was about Out Of Focus Image Restoration (Stills Images). I just said that if you estimate the PSF (The Blurring Function) well you get a great results from Deconvolution.
It's not easy, yet it restores details no Sharpener can.

Theoretically Super Resolution should yield better results since you have not only one frame but more. If you utilize that into a good Deconvolution Algorithm you should get great results on details that were blurred.

I don't know about the Plug In's the OP wrote.
Yet if Didee rights about the impressions he has, I guess the Super Resolution Algorithm implemented there can't get what theoretically it might.

Anyhow, one day maybe someone will make such Plug In.

[zee944]

Originally Posted by royia

Did you read my messages?
Never used it.

Mmm, sorry, I thought you've written the original post too.

[Hagbard23]

"Deconvolute" THIS:

http://forum.doom9.org/showthread.php?t=157047

That should show what our opinion here is. If you have a video without details, which is so far out "flat" you can't do anything - even with your super-hyper-mega-scientific "Super-Resolution" Filter. It is just impossible to restore something, that is not there anymore. Maybe you get to the point, that you can restore some single pixels, but you have to invest hours and hours for the calculation - and if i understood it right, you have to do a new calculation for every new video source.

I a time of HD_content present all over the place, noone sees the sense of spending so much energy for calculating things, you can get easily -without work- if you choose the right source. And if you have video, which -for some reason- is not available in HD (old b/w films for example) you will not get any gain in details, because of the noise all over the place - and so it is not worth the time. The best, you can achieve with those sources is to amplify noise.

It's not, that we don't respect the whitepapers, which -in theory- are making sense. It is, that we don't see efficiency and practicability. Its not, that you write such a plugin in some hours or days. You really have to invest time and work - for -lets say- nearly nothing.

Sorry, if my statement was some little unkind. It was not meant personally...

[royia]

Hello.
No Problems.

I have no Video Experience.
As I wrote it can be done for Still Frames.
I'm not talking about restoring an image which was blurred by something equivalent of Gaussian Blur with STD of 10.
There is no come back from there.

My algorithm worked well for things equivalent of GB with STD about 3 and below.
That's something.

All I said that in Video you have more data.
If you use it well you can better result that I got.
Moreover, I'm sure there are out there people with much more experience than I have about Image Restoration.
So even better results are achievavle.

You're right about speed.
My algorithm takes about 3 sec per channel per frame (I worked on Grayscale Images of size 800x800).
It's not meant for HD that's for sure.

I just there's something decent to make a 720x576 Video look good even though it was captured by a 270,000 Pixels CCD.

[JClement]

Go to the Smooth Video Project (SVP)
https://www.svp-team.com/wiki/Main_Page

They have free AviSynth plugins that do, to my eyes, a perfect job of frame rate conversion. Neither of the free MSU FRC plugins do anywhere near as good a job. I used SVP to double the frame rate of a 16fps 8mm film to 32 and set the final frame rate down to 30. I have not tried other ratios, but it can be set to any input/output ratio you want. Doubling the frame rate with it was much faster than the MSU plugin. The older free MSU left horrible artifacts with tall thin objects broken into horizontally displaced blocks during pans. The newer demo cut off the tops of some tall thin objects on alternate frames. SVP just looks perfect either when run up to speed or single stepped.

The SVP plugin pack includes Apple, Windows, an Linux plugins, so you have your choice. Here is a sample AVS script. I commented out the parts that didn't work.
--------------------------------
SetMemoryMax(1024)
source = AVISource("C:\folder...\file.avi").ConvertToYV12()
source
LoadPlugin("C:\Program Files (x86)\AviSynth 2.5\plugins\svpflow1.dll")
LoadPlugin("C:\Program Files (x86)\AviSynth 2.5\plugins\svpflow2.dll")

threads=9
#SetMTMode(3,threads)
# Some input here
#SetMTMode(2)

# All parameters set to defaults which means high quality frame doubling
super=SVSuper("{gpu:1}")
vectors=SVAnalyse(super, "{}")
SVSmoothFps(super, vectors, "{}",mt=1)
-----------------------------------
The parameter for rate change is:
Rate:{Num=2, Den=1, abs=false}
These are the numberator, denominator, and abs=true means absolute frame rate rather than a multiplier of the original rate.

[JClement]

A VirtualDub filter Deshake does a good job.
http://www.guthspot.se/video/deshaker.htm
It is not free, but the cost is very reasonable. While the AvisSynth filter may work as well, the ability to preview and easily change parameters is valuable.

Many video editors such as Magix Movie Edit Pro also have deshakers, but with limited options. The MEP deshaker gives greater stability when adjusted well, but can't retain the original full frame size. Deshake can retain the original full frame by using earlier and later frames. However this has problems with moving objects at frame edges and progressive shifts in brightness. The large number of options can seem daunting, but with a bit of playing it can produce excellent results. However you can not use it with other filters that use more than one frame at a time such as Temporal smoother.