modPlus plugin for avisynth+

[vcmohan]

modPlus , a new plugin to work with avisynth+ r2347 is now available for down load on my plugins page. There are 9 functions in this. All are thread safe and accept all bit depths and float inputs. Restrictions if any, are mentioned in the detailed description which can be seen by clicking the name above. The functions were available as separate plugins for avisynth 2.5, 2.6 having slightly different names.

Functions:
1.Median
2.GBlur
3.MBlur
4.minvar
5. Morph
6.SaltPepper
7.SegAmp
8.TweakHist
9.Veed
10 Fan
11 Neural

These functions modify the pixel values on different criteria to suppress noise, or aid in suppressing noise in conjunction with other functions / plugins. All these are now in one plugin, hence the name. While upgrading some parameters were altered from the original versions and hence changed names.

Bug reports and suggestions for additions are welcome.

[vcmohan]

A new function Morph has been added to this plugin. Using morphological principles image can be dilated or eroded or opened or closed. In addition a median filtering function is also included. Use of this median filter as a pre processor to dilate or erode or open or close is preferable.

[vcmohan]

I have now added two functions : Fan and Neural to this plugin. Fan is improved version of FanFilter (of 2.5) and works better. I have posted a script in a thread of eliminating single frame white streaks. This can also be used for attenuating regular freq noise and random noise.
Neural is a ANN function. Given input and a target output having some unknown linear relationship, it learns through a number of iterations and results look good. Recently posted @ingoldie thread Original and result videos I used with good result.

[hello_hello]

Thank you!

[real.finder]

hi
mblur and some others in x64 seems crash in avs+ 3.6.2 (at least https://github.com/AviSynth/AviSynthPlus/issues/190),

BlankClip(pixel_type="yv12") 
modPlus_mblur

Originally Posted by vcmohan

Neural is a ANN function. Given input and a target output having some unknown linear relationship, it learns through a number of iterations and results look good. Recently posted @ingoldie thread Original and result videos I used with good result.

Neural seems crash on me in both x64 and x86, can you please post new usage example?

[vcmohan]

I am consolidating modplus, moveplus, fqplus and some other into one plugin. In a few days I will upload it. I will test all of the functions as I am also updating them. Just bear with me. I am a bit slow.

[real.finder]

thanks, I will waiting for it, and BTW I note that the Neural in modplus seems missing many things that was in old http://www.avisynth.nl/users/vcmohan...NeuralNet.html

[real.finder]

seems I use old modplus for x64, now mblur work after using last modplus

but still can get Neural to work

ColorBars(width=640, height=480, pixel_type="yv12")
Neural(Blur(1), iter = 3000 , wset = false, bestof = 1,tty = 100, tlx = 100, trx = 300, tby = 300, xpts = 7)

give

[vcmohan]

Originally Posted by real.finder

but still can get Neural to work

I just checked with this script and looked works fine

a=colorbars()
b = GBlur(a,rad = 7, sd = 2.0)
#a=converttoRGB24(a)
#b=converttoRGB24(b)
Neural(a,b, iter = 1000, bestof = 6, wset = false,tty = 100, tlx = 200, trx = 400, tby = 300, xpts = 7, txt = "save", fname = "C:\temp\neuralfile.txt")
stackhorizontal(a,last,b)

It takes time to train, looks as though hanging, so be patient. In this example I am using blurred frame as the target. Of course it is silly, but shows working.

[real.finder]

Originally Posted by vcmohan

I just checked with this script and looked works fine

a=colorbars()
b = GBlur(a,rad = 7, sd = 2.0)
#a=converttoRGB24(a)
#b=converttoRGB24(b)
Neural(a,b, iter = 1000, bestof = 6, wset = false,tty = 100, tlx = 200, trx = 400, tby = 300, xpts = 7, txt = "save", fname = "C:\temp\neuralfile.txt")
stackhorizontal(a,last,b)

It takes time to train, looks as though hanging, so be patient. In this example I am using blurred frame as the target. Of course it is silly, but shows working.

yes, that work, so what is the problem with my previous use?
edit: I think I know, seems the train clip is not used (it use it as input one instead of last) and seems it's not optional

and is it possible to train using another frame not the 0 one? and also can we use the data in neuralfile without forced to use train clip?

[vcmohan]

If you want to use another frame as tclip, you can always trim that frame as tclip = trim(last, n, n) and use it. The read option can be used to read the file back next time to save time. However the formats should be same and the saved file was not edited.

[real.finder]

Originally Posted by vcmohan

If you want to use another frame as tclip, you can always trim that frame as tclip = trim(last, n, n) and use it. The read option can be used to read the file back next time to save time. However the formats should be same and the saved file was not edited.

I already know that, but I mean in next update it will be nice if there are parameter to chose the frame like the old plugin

also I mean in read option the train clip should not be needed (I tried without it using read option and still get the access violation)

[vcmohan]

Originally Posted by real.finder

I already know that, but I mean in next update it will be nice if there are parameter to chose the frame like the old plugin

also I mean in read option the train clip should not be needed (I tried without it using read option and still get the access violation)

OK. I will include your suggestions in the newer consolidated version 'manyPlus', It is ready but I need to update the documentation prior to uploading. May take some time.