Degraining script assist?

[johnmeyer]

Originally Posted by Bruce/Fl

How much (length) of a sample would be best? Taking file size into consideration. The sample scan they sent me was approximately 10 minutes. I would assume you would only need 10, 15 seconds worth?

5-10 seconds, preferably something with motion (i.e., static shots of a building are not as useful as something where people are walking or cars are driving by, etc.).

[Bruce/Fl]

This should be about 6 seconds of my sample. Here is the link. I used Vegas, rendered it as .m2ts

https://www.dropbox.com/s/wpgso9p8aw...load.m2ts?dl=0

[johnmeyer]

Wow, that is really grainy. As a result, any attempt to use VideoFred's script will result in a fine mess because his scripts do quite a bit of sharpening. Sharpening grain, even after "degraining" is going to produce bad results. In fact, this is one of the main things that drove me to create my version of his script. If you click on the link below, you can see my post in his original film restoration thread -- complete with several images -- that shows the problems his original script created with my grainy footage:

The power of Avisynth: restoring old 8mm films

Also, when you actually do the work, make sure to work with the original MOV. This sample you uploaded has been recompressed using the Sony AVC codec, and there are a lot of additional artifacts added to the grain.

Finally, if this clip really accurately represents the original (i.e., you didn't screw up when you rendered it), then you need to decimate half the frames before you do any degraining. This is because the frames are all duplicated, and the degraining will choke when it sees zero motion between half of the frames.

Here is the script I just used on your clip. I frameserved it from Vegas, which is what I always do in order to avoid having to figure out how to read dozens of different formats into AVISynth. I let Vegas do that job because uses professional codecs that seldom glitch or fail. Note the "selecteven()" statement in the first line. This gets rid of every other frame, thus eliminating all the duplicates. The Assumefps statement gets you back to the correct playback speed.

So, you will need to replace the AVISource() statement with whatever you have found works for getting your MOV files into AVISynth. As I remember from our PMs last week, you did figure that out and got it working.

Hopefully other people will provide some additional ideas. I'm sure they will be glad to critique my work

source=AVISource("e:\fs.avi").selecteven()
super = MSuper(source,pel=2, sharp=1)
backward_vec2 = MAnalyse(super, isb = true, delta = 2, overlap=4)
backward_vec1 = MAnalyse(super, isb = true, delta = 1, overlap=4)
forward_vec1 = MAnalyse(super, isb = false, delta = 1, overlap=4)
forward_vec2 = MAnalyse(super, isb = false, delta = 2, overlap=4)
MDegrain2(source,super, backward_vec1,forward_vec1,backward_vec2,forward_vec2,thSAD=1400)
final=assumefps(15/1.001)
return final
#stackhorizontal (source,final)

[Bruce/Fl]

Originally Posted by johnmeyer

This sample you uploaded has been recompressed using the Sony AVC codec, and there are a lot of additional artifacts added to the grain.

Does anyone here know if Vegas can trim a .mov clip losslessly, or if not, what is the best option for rendering a .mov file? I tried the quicktime 7 (.mov) options, but even to my eye they looked very splotchy. (<---- ya, my word)

Or possibly some other program? I'd love for these samples I'm uploading to be simply shorter versions with the same quality as what I received.

Thank you,

Bruce

[johnmeyer]

Use Mediainfo or similar tool to find out what codec is used by the MOV file. Vegas may give this to you if your right-click on the media and select "properties."

In general, Vegas does not "smart render" (a.k.a., lossless cuts) most video. It can smart render DV AVI and some forms of MPEG-2, but that's about it.

Having said that, if the version of ProRes (I think that's what you said this is) does not do interframe compression, Vegas may well be able to render it losslessly.

It is easy to tell: just start the render and watch the preview screen in Vegas while rendering. If the video is not being re-rendered, the display will show "no compression needed."

Did my script work for you?

[Bruce/Fl]

Originally Posted by johnmeyer

Use Mediainfo or similar tool to find out what codec is used by the MOV file. Vegas may give this to you if your right-click on the media and select "properties."

In general, Vegas does not "smart render" (a.k.a., lossless cuts) most video. It can smart render DV AVI and some forms of MPEG-2, but that's about it.

Having said that, if the version of ProRes (I think that's what you said this is) does not do interframe compression, Vegas may well be able to render it losslessly.

It is easy to tell: just start the render and watch the preview screen in Vegas while rendering. If the video is not being re-rendered, the display will show "no compression needed."

Did my script work for you?

No, it said "Avisynth open failure, there is no function named "MSuper" "

I also linked it to a .avi render I did, so that it would work with what you gave me as written.

Attached are the media info screen caps from the original .mov and also my AVC render, just because.

Originally Posted by johnmeyer

This is because the frames are all duplicated, and the degraining will choke when it sees zero motion between half of the frames.

I was curious about this, after reading your explanation about how the degraining process worked, when I suspected that I did have duplicate frames. The reason I suspected that was, when I was trying to find the first frame of a scene in Vegas, using the frame advance button, it always seemed like I needed to click it twice to see any change in the image.

[johnmeyer]

Originally Posted by Bruce/Fl

No, it said "Avisynth open failure, there is no function named "MSuper" "

You have to put the MVTools2 DLL in your AVISynth plugins folder.

[Bruce/Fl]

Originally Posted by johnmeyer

You have to put the MVTools2 DLL in your AVISynth plugins folder.

It is now in there, now it says: "MSuper: Clip must be YV12 or YUY2

Attached is the media info on the avi. It's huge, it's 4:2:2 and it says it's lossless.

I can probably uplaod it if you'd like to see it, better quality than the m2ts maybe?

[StainlessS]

Bruce/Fl,

You can just copy and paste MediaInfo text into post, click 'Go Advanced', and then select text and click on the Hash symbol above the post entry box, it can take a long time for attachments to clear approval.

You can see what colorspace Avisynth thinks clip is by pasting

Return Info.ConvertToRGB32

immediately after the line that loads the clip.

[raffriff42]

And here's some help for cutting down that huge source file.

[Bruce/Fl]

Originally Posted by raffriff42

And here's some help for cutting down that huge source file.

That worked great, thank you so much for that.

Now I'll upload a few seconds of the original, for anyone wanting to see it un-reencoded.

Bruce

[Bruce/Fl]

Here is a sample in the original format, .mov


https://www.dropbox.com/s/2xk8v5r36e...8mm_1.mov?dl=0

[Bruce/Fl]

Originally Posted by StainlessS

Bruce/Fl,

You can see what colorspace Avisynth thinks clip is by pasting

Return Info.ConvertToRGB32

immediately after the line that loads the clip.

I added it. Now it says, "I don't know what "info" means"

And just to be sure there are no mistakes I've made, here is what I have writtin:

source=AVISource("C:\Users\Bruce\Desktop\upload.avi").selecteven()
Return Info.ConvertToRGB32
super = MSuper(source,pel=2, sharp=1)
backward_vec2 = MAnalyse(super, isb = true, delta = 2, overlap=4)
backward_vec1 = MAnalyse(super, isb = true, delta = 1, overlap=4)
forward_vec1 = MAnalyse(super, isb = false, delta = 1, overlap=4)
forward_vec2 = MAnalyse(super, isb = false, delta = 2, overlap=4)
MDegrain2(source,super, backward_vec1,forward_vec1,backward_vec2,forward_vec2,thSAD=1400)
final=assumefps(15/1.001)
return final
#stackhorizontal (source,final)

[videoFred]

Originally Posted by Bruce/Fl

It is now in there, now it says: "MSuper: Clip must be YV12 or YUY2

Avisynth is telling you that MSuper requires YV12 or YUY2 colorspace. How to fix this? By changing color space from the input of cource.

Change your first line like this:

source=AVISource("C:\Users\Bruce\Desktop\upload.avi").selecteven().converttoYV12()

Fred.

[Bruce/Fl]

Originally Posted by videoFred

Avisynth is telling you that MSuper requires YV12 or YUY2 colorspace. How to fix this? By changing color space from the input of cource.

Change your first line like this:

source=AVISource("C:\Users\Bruce\Desktop\upload.avi").selecteven().converttoYV12()

Fred.

Thanks Fred, that got rid of the trouble boxes, and I have an image in virtualdub now. A very large image, but an image, and it moves, slowly.

Unfortunately I have to run now, but will return to this later today and see what I can do/see.

Again, thanks all.

Bruce

[Boulder]

Based on my experiences, you definitely want to preclean the clip you analyse motion from if the amount of noise or grain is extreme. I'd try FFT3DFilter or some of those NLMeans-type denoisers.

[StainlessS]

Originally Posted by Bruce/Fl

I added it. Now it says, "I don't know what "info" means"

And just to be sure there are no mistakes I've made, here is what I have writtin:

source=AVISource("C:\Users\Bruce\Desktop\upload.avi").selecteven()
Return Info.ConvertToRGB32

Arh, OK, you assigned clip to source variable, so I should have posted:

return source.info.ConvertToRGB32

instead.
I assumed that you have not assigned to any named variable and so would have been assigned to the special 'implied' Last clip variable.

What I originally posted was equivalent to this:

AVISource("C:\Users\Bruce\Desktop\upload.avi").selecteven()  # Implied assignment to special Last variable.
return Last.info.ConvertToRGB32
# return info.ConvertToRGB32              # EDIT: Or use implied Last without explicit naming

JFYI.

[mawen1250]

About thSAD, from what I have read before(just a glance at the mvtools code):
It actually affacts the denoising strength of MDegrain, since MDegrain weighted average the matched blocks, and the weighting is calculated from SAD and thSAD. Thus, a higher thSAD results in dissimilar blocks to have larger weight, and it can increase denoising strength as well as introduce more blending artifacts.
As for MCompensate based denoising, thSAD might just be a threshold to determine bad matches.

[feisty2]

inline float DegrainWeight(double thSAD, double blockSAD) {
    if (blockSAD >= thSAD)
        return 0.f;

    return float((thSAD - blockSAD) * (thSAD + blockSAD) * 256 / (thSAD * thSAD + blockSAD * blockSAD));
}

if (blockSAD >= thSAD)
return 0.f;

threshold

weight = 256 * (thSAD^2 - blockSAD^2) / (thSAD^2 + blockSAD^2)
you got thSAD at both multiplication and division part, who knows what will happen to the weight if you increase it, not doing a monotonicity determination of "y = (x^2 - a)/(x^2 + a)" here

[feisty2]

well... I just did it anyways, cuz that's just some 15 years old simple math
turns out "y' = 4ax / (x^2 + a)^2", and y' always > 0
so, guess mawen1250's right, you increase the weight by increasing thSAD

EDIT:
but that doesn't really matter, you know, it's an "x/x^4" kinda thing... it grows really, really, really, slow, yeah, you know what I mean....