MVTools-pfmod

[pinterf]

I like when the questions get answered without me. btw masks, I think I still have some work on the mask side for 10+ bits. Are masks 8 bit in the vapoursynth mvtools version?

[feisty2]

there's still no high bit depth support for MMask in jackoneill's branch, but I got my 32bits floating point version of MMask in my branch.
https://github.com/IFeelBloated/vapo...src/MVMask.cpp

[real.finder]

I think there are something related to mvtools here https://forum.doom9.org/showpost.php...postcount=2286

cuz QTGMC(SLMode=0,tv_range=false,tr1=0,tr2=0)

will be ok in all bit depth

[pinterf]

Originally Posted by real.finder

I think there are something related to mvtools here https://forum.doom9.org/showpost.php...postcount=2286

cuz QTGMC(SLMode=0,tv_range=false,tr1=0,tr2=0)

will be ok in all bit depth

What difference should I recognize? Yes, there are minor differences, but only when I'm looking at the same frames with the alt-tab method and pixel-peeping. I'm testing with a Hi8 DV-AVI source.

[real.finder]

Originally Posted by pinterf

What difference should I recognize? Yes, there are minor differences, but only when I'm looking at the same frames with the alt-tab method and pixel-peeping. I'm testing with a Hi8 DV-AVI source.

it look sharp and somehow with some artefact

[real.finder]

even in SMDegrain, so I think there are something in MDegrainX

and something else, can you add error message for not use lsb with native high bits?

[pinterf]

Can you send me a short clip sample where the differences are well visible? It would save me time to have a good test case, thank you in advance

[real.finder]

Originally Posted by pinterf

Can you send me a short clip sample where the differences are well visible? It would save me time to have a good test case, thank you in advance

done in pm

[pinterf]

New release with an important fix for 10+ bits
MvTools2 2.7.15.22 with depans

MvTools2 2.7.15.22 (20170316)
Fix: 16 bit SAD for non-AVX code path
Misc: MDegrain1-6: add error on lsb_flag=true for non-8 bit sources

[MysteryX]

Here's a bug with MVTools2 with YV24.

Here's what it looks like on a still video, but I don't think it matters which source you use.

AviSource("Preview.avi")
ConvertToYV24()
jm_fps(60)


function jm_fps(clip source, float "fps")
{
fps = default(fps, 25.000)
fps_num = int(fps * 1000)
fps_den = 1000

prefiltered = RemoveGrain(source, 22)
super = MSuper(source, hpad = 16, vpad = 16, levels = 1) # one level is enough for MRecalculate
superfilt = MSuper(prefiltered, hpad = 16, vpad = 16) # all levels for MAnalyse
backward = MAnalyse(superfilt, isb = true, blksize = 16, overlap = 4, search = 3, dct = 0)
forward = MAnalyse(superfilt, isb = false, blksize = 16, overlap = 4, search = 3, dct = 0)
forward_re = MRecalculate(super, forward, blksize = 8, overlap = 2, thSAD = 100)
backward_re = MRecalculate(super, backward, blksize = 8, overlap = 2, thSAD = 100)
out = MFlowFps(source, super, backward_re, forward_re, num = fps_num, den = fps_den, blend = false, ml = 200, mask = 2)

return out
}

[pinterf]

Thanks, reproduced. YV16 and YV24 both have artifacts
EDIT:
Could you send me a link to your source?
Because your script produces the same artifacts for all colorspaces YV12, YUY2, YV24 for me.
Checked YV12 and YUY2 with older mvtools versions 2.5.11.3, 2.5.11.22, 2.6.0.5 and they are also ugly.

[MysteryX]

I often use this source for testing; with LSMASHSource
https://www.spiritualselftransformat...ncoder-old.mpg

[pinterf]

Originally Posted by MysteryX

I often use this source for testing; with LSMASHSource
https://www.spiritualselftransformat...ncoder-old.mpg

Thanks. Still I can see no differences between YV12/YV24 or even YUV444P16.
Are you using 2.7.15.22?
If so, what processor capabilities do you have? (Different code paths?)

[MysteryX]

Originally Posted by pinterf

Thanks. Still I can see no differences between YV12/YV24 or even YUV444P16.
Are you using 2.7.15.22?
If so, what processor capabilities do you have? (Different code paths?)

I'll be traveling until the 12th. I can do some more testing later if you haven't figured it out yet.

[Motenai Yoda]

Originally Posted by Motenai Yoda

depanestimate (or depan) parameters have been changed? coz scene change detection looks broken to me

It isn't fixed yet?

[real.finder]

Originally Posted by Motenai Yoda

It isn't fixed yet?

it is

Depan 2.13.1.2 (20161228)
Fix: DepanStabilize: removed too strict checking for large motion vectors received from depan
that resulted in like scene change

[Motenai Yoda]

Originally Posted by real.finder

it is

I got it,
but what I was/am experiencing is a lot of scene-changes not detected...

[VS_Fan]

Originally Posted by Motenai Yoda

I got it,
but what I was/am experiencing is a lot of scene-changes not detected...

Have you tried using lower values for the thSDC2 parameter?
This is what I usually do: The default for thSCD2 is 130 equivalent to the 51% of blocks changed. Try lower values in steps of 8, which corresponds approximately to steps of 3%: to 128, 120, 112, etc.

You can use showSAD parameter of the mShow function to estimate both thSCD1 & thSCD2 values:

return mshow(super, vectors, showsad=True, thSCD1=400, thSCD2=112, sil=1)

After determining the appropriate values of thSCDx paramenters, you can use the info=True parameter of the mDepan Function, to refine them more precisely:

globalmotion = mDepan(pf, vectors, zoom=True, pixaspect=1.0, thSCD1=400, thSCD2=112, info=True, error=60.0, wrong=40.0)
return globalmotion

BTW: error and wrong parameters of mDepan can be raised a bit from the defaults to make it more tolerant to big movements that could otherwise reset the motion compensation as if it were due to Scene Change

Finally, you can use the info=True parameter in the DePanStabilize function to verify the results.

And don't forget to comment out the whole mShow function and set to False the showsad and info parameters

[Motenai Yoda]

I'm not using mdepan but depanestimate

s = 8
stab_reference = removegrain(4,-1,-1).mt_edge(mode="min/max",U=-128,V=-128)
data = depanestimate(stab_reference,trust=2.0,dxmax=s,dymax=s,stab=12,zoommax=1.0)
DePanStabilize(cutoff=1.0,data=data,dxmax=s,dymax=s,zoommax=1.0,rotmax=0.0,mirror=15,blur=30,method=1,subpixel=2)

[pinterf]

Ok, I'll make a comparison with this script