Strange fields/frames in analog capture

[lollo]

I have an analog capture where the frame and field patterns is weird: a mixed of interlaced frames and progressive frames that I know how to manage, but then blends at (even) field level.



Any idea how to deal with the last?

[Sharc]

Try

yadifmod2(mode=1,order=1)
srestore(frate=25/1.000)

[jagabo]

There nothing unusual about that at all. It's a standard field blended NTSC to PAL conversion. As Sharc points out, use a smart bob followed by SRestore. I think the frame rate should be 23.976 though.

[lollo]

Thanks to both, I will try what you suggest

[jagabo]

LWLibavVideoSource("spots_beg_ufo_sIII1c_amtv_v2_cut.avi") 
AssumeTFF()
QTGMC()
SRestore()

[Selur]

@jagabo: which QTGMC and sRestore are you using?
When I use:

# loading source: C:\Users\Selur\Desktop\spots_beg_ufo_sIII1c_amtv_v2_cut.avi
LWLibavVideoSource("C:\Users\Selur\Desktop\SPOTS_~1.AVI")
AssumeTFF()
QTGMC()
srestore()
return last

I get different results than your clip, mine still got ghosting.
I suspect that something is wrong with my QTGMC or sRestore version.
(probably QTGMC)
(I attached the versions of the script that I used here.)

Cu Selur

[lollo]

I get different results than your clip, mine still got ghosting.

Same for me, using QTGMC v3.364 and srestore v2.797 I still have ghosting.


Edit: adding images

my QTGMC+Srestore version, now also odd field has blending


jagabo QTGMC+Srestore version

[jagabo]

I have QTGMC v3.375 and Srestore v2.796. But I tried using the versions Selur uploaded and got the same result as my earlier upload.

Keep in mind that SRestore may take a while to lock into the clear/blend pattern. If you just seek around in an editor it may not give clean results. Seek about 30 frames back and step forward frame by frame until you get to the frame you want to check.

[Sharc]

QTGMC produced 2 or 3 back/forth jerks and few residual ghosts. That's why i suggested yadifmod2 as the bobber, which did solve the issue here. Don't know why yadifmod2 behaved better.
My QTGMC is 3.382s and SRestore is 2.795.

[jagabo]

QTGMC produced 2 or 3 back/forth jerks and few residual ghosts. That's why i suggested yadifmod2 as the bobber, which did solve the issue here. Don't know why yadifmod2 behaved better.
My QTGMC is 3.382s and SRestore is 2.795.

Did you set the field order before calling QTGMC? AssumeTFF().

[Sharc]

QTGMC produced 2 or 3 back/forth jerks and few residual ghosts. That's why i suggested yadifmod2 as the bobber, which did solve the issue here. Don't know why yadifmod2 behaved better.
My QTGMC is 3.382s and SRestore is 2.795.

Did you set the field order before calling QTGMC? AssumeTFF().

Worse even. I still had AssumeBFF() in the script from a leftover. My bad.
It's ok now with AssumeTFF().
I don't think SRestore can completely and perfectly remove the ghosts, but it's as good as it can get.

[Selur]

Keep in mind that SRestore may take a while to lock into the clear/blend pattern. If you just seek around in an editor it may not give clean results. Seek about 30 frames back and step forward frame by frame until you get to the frame you want to check.

okay, that did it for me in Avisynth.

[lollo]

okay, that did it for me in Avisynth.

That did it for me as well. Once the output of the AviSynth script is fully created through VirtualDub, it is ok.

Thanks jagabo for the hint, the temporary visualization of the script with VirtualDub is not accurate. A good lesson for the future.


[Attachment 67114 - Click to enlarge]

[johnmeyer]

This same hint applies to quite a few other AVISynth filters.

It gets even more complicated if you use any sort of AVISynth multi-threading (different versions of AVISynth do this differently). In some cases I cannot do any seeking and have to disable all multi-threading in order to test the scripts. Then, the only way some scripts will work correctly is if they are started at frame 0, and then only go forward from that point.

[lollo]

This same hint applies to quite a few other AVISynth filters.

Thanks John. Yes, I experienced that, for example, a temporal denoiser with high temporal radius (i.e. 12) needs seeking through many frames to show its results while testing a script. I did not play attention to Srestore behaviour, then my mistake.

if you use any sort of AVISynth multi-threading

I never did and I will never do

[Selur]

Hmm,... seems like adding a

Preroll(Int(last.FrameRate()))

behind the sRestore call also seems to fix the issue.

[jagabo]

Hmm,... seems like adding a

Preroll(Int(last.FrameRate()))

behind the sRestore call also seems to fix the issue.

Nice find. I didn't know there was a Preroll filter built into AviSynth+ now. In the past I've used TIVTC's RequestLinear() filter for that.

[Selur]

iirc. Preroll was added with Avisynth 2.6 even before Avisynth+ (but I could be wrong )