Flying erase head noise

[lordsmurf]

I need the rainbows removed. I'm not liking what I'm doing, it still has rainbow bits.

What are your ideas here?

[Alwyn]

I'll avago. Neat.

[themaster1]

well hard to test something with such a odd frame size 220 x339

try this

AVISource("C:\flying era head noise.avi")
assumetff()
addborders(0,0,0,1)
#spline36resize(404,last.width)
ConvertToYV16(interlaced=true)
orig=last
ev=orig.assumetff().separatefields().selecteven()
od=orig.assumetff().separatefields().selectodd()
ev
ue_chroma = UToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
ve_chroma = VToY(ev).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(ue_chroma, ve_chroma)
MergeLuma(ev)
ev_filtered=last
od
uo_chroma = UToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
vo_chroma = VToY(od).blur(0,1.5).ttempsmooth(maxr=6,lthresh=15 0, strength=6).KNLMeansCL(d=3,a=8,s=2,h=6)
YToUV(uo_chroma, vo_chroma)
MergeLuma(od)
od_filtered=last
interleave(ev_filtered,od_filtered)
assumefieldbased().assumetff().weave()

[Alwyn]

Themaster, I'm getting this on your script:


[Attachment 78540 - Click to enlarge]

[Selur]

Spotless only applied to chroma probably works.

[johnmeyer]

I had good luck with Spotless when I used it on chroma rainbows. Scroll down to my post (#8) to see the before/after, along with the script I used.

https://forum.videohelp.com/threads/398864-Need-help-on-Avisynth#post2595644

[Alwyn]

Not bad, John.

[lordsmurf]

Alwyn, thanks for trying. This task is probably beyond what NeatVideo can do.

__________________

john, I get this on your script



line 3 is

output = SpotLess(ThSAD=1000,ThSAD2=1000,RadT=4,Chroma=true,BlkSz=8,OLap=4,Tm=true)

My understanding is that was more for rainbowing from composite, not as much head-switching type errors.

__________________


themaster1, and oldie but goodie, I forgot to try that, thanks.

Well, only 3 hours left for a 6 minute clip.
But I'm running it now.

[Alwyn]

LS, I got a warning about being divisible by 4, so I cropped it to 336. Might be the same issue?

[lordsmurf]

Alwyn, I tried running it on the full 720x480 clip.

themaster1, that did the trick.

I have several KNLmeansCL scripts in my toolbox, forgot about that one It took forever, about 4.5 hours for 6 minutes of footage. I probably could have lowered some of the values, for better speed, but I needed to get off the computer for a while anyway. Let it crunch!

@all, appreciate it.

However, I'm still curious why john's script won't load.

[DTL2023]

With latest release of vsTTempSmooth with 'large radius median' mode - https://github.com/Asd-g/AviSynth-vsTTempSmooth/releases/tag/1.2.6

LoadPlugin("ffms2.dll")
LoadPlugin("vsTTempSmooth.dll")

FFMpegSource2("flying era head noise.avi")
assumetff()
addborders(0,0,0,1)
noproc=last

ConvertToYV16(interlaced=true)
orig=last

ev=orig.assumetff().separatefields().selecteven().vsttempsmooth(maxr=10,uthresh=55, vthresh=55, y=2, u=3, v=3, pmode=1, opt=0)
od=orig.assumetff().separatefields().selectodd().vsttempsmooth(maxr=10,uthresh=55, vthresh=55, y=2, u=3, v=3, pmode=1, opt=0)
interleave(ev,od)
assumefieldbased().assumetff().weave()

StackHorizontal(noproc.Subtitle("input"), last.Subtitle("proc"))

Though I not sure what will happen on moving areas with such large setting for threshold - as usual it is recommended to process motion-compensated only sequences with such temporal methods (use MCompensate to create MC sequence and SelectEvery to pick center processed).

[johnmeyer]

My script is only five lines (the rest is just the Spotless code). Looking at it, I used implicit variables from one line to the next, so I think the only thing which would cause an invalid argument call in line 4 (the line which calls Spotless) is from the file open statement on line 1.

You obviously need to use your own file name instead of "fs.avi" which is the file server stub that I use to serve out of Vegas and into AVISynth. You also may need to use something other than AVISource, depending on the format of your original video file. You can test this out by temporarily changing "Return Output" to "Return Source." That reduces it to a two-line script. If the script is having a problem reading the file, you'll still get an error.

I don't think the script cares whether the rainbowing happened from head switching or something else. The script acts on what it sees, not on what caused it. However, since there is a luma component as well, Spotless will probably only remove the rainbow, although if you turn up the settings high enough, it might get rid of the vertical bars as well.

[rgr]

Edit: Ah, it's something different than I thought

I need the rainbows removed. I'm not liking what I'm doing, it still has rainbow bits.

What are your ideas here?

Check this thread:
https://forum.videohelp.com/threads/413191-How-to-remove-horizontal-lines-noise

especially this post:
https://forum.videohelp.com/threads/413191-How-to-remove-horizontal-lines-noise#post2721152
+ TemporalDegrain2 (or Neat)

[Gelinox]

And without Avisynth ?

[Selur]

@lordsmurf: try, applying SpotLess with temporal radius 3, 3 times in a row (and take the chroma from that), it's not perfect, but seems like a good start.

Cu Selur

[DTL2023]

Median filtering with temporal radius of 3 frames reduces defect but not clean completely.

I think it is equal to
ev=orig.assumetff().separatefields().selecteven(). vsttempsmooth(maxr=3,uthresh=55, vthresh=55, y=2, u=3, v=3, pmode=1, opt=0)
od=orig.assumetff().separatefields().selectodd().v sttempsmooth(maxr=3,uthresh=55, vthresh=55, y=2, u=3, v=3, pmode=1, opt=0)

Spotless() is a motion-compensated version of the same processing (but MedianBlurTemporal in Spotless() run as a median filtering without any limiting of difference between filtered and current sample value). So I recommend the authours of Spotless() replace MedianBlurTemporal with new vsTTempSmooth or may be add it as an option because MedianBlurTemporal is faster at special cases of tr=1 and 2 (?).

Also in the description of 3-frames temporal median Clense() it was noted running several times equal to temporal median with larger tr (initial tr or Clense is 1 for 3 frames total). It may be also tested.

"applying SpotLess with temporal radius 3, 3 times in a row"

Can you check single run of Spotless() with RadT of 10 (I do now see the limits of temporalradius in the MedianBlurTemporal at http://avisynth.nl/index.php/MedianBlur2) ?

[Selur]

Can you check single run of Spotless() with RadT of 10

attached

[DTL2023]

What mean 3xSpotless in filename ? Is it an error ? Is it a single call to Spotless with RadT 10 only ?

*after updating ffms2.dll to latest RC5.0 to read that strange AV01 format* - yes - it looks clear enough too,

[Selur]

Sorry, that that was a mistake during the file naming, only applied Spotless with radT 10 once in that file.