soft pulldown but still interlaced?

[m3mbran3]

I have a NTSC DVD boxset of a TV series "freaks and geeks" that I am trying to encode with x264.

If I review the .vob in DGIndex then I get a framerate of 29.97fps, a video type of Film 94.60% and many Field Rpts. From reading around this seems to indicate that the source is actually 23.97fps with a soft pulldown applied. So I save my .d2v with Force Film applied and thought that my content should be all progressive now.

When previewing the file with vdubmod it appeared as if most of the content was progressive but there were still a few interlaced frames appearing when a camera angle was changed.

i.e. P-P-P-P-P-P-P-P-P-P-I-I-[scene change]-I-P-P etc.

This happens at about half of all camera angle changes and I can only assume this is a bi-product of the video editing process.

If I then load this .d2v file into megui and run the video analysis it informs me that content is partially interlaced with a varying field order and adds "Yadif(order=-1)" to my .avs file.


If I run the same input .vob through DGIndex with ignore pulldown checked then I get the same results as before except with no Field Rpts and the resulting .d2v is 29.97fps as opposed to the force films 23.97fps.

This .d2v file when run through the source detector of megui says that the source is film with top field first and adds the avs line "tfm(order=1).tdecimate()"


I have previewed both scripts in vdubmod and both get rid of interlacing and play at 23.97fps. I was just wondering what the most correct way of doing this was. At the moment I am just using what megui gives me which is exactly the same as the 2nd method and the output video appears to be of acceptable quality.

[Inspector.Gadget]

Ignore pulldown is not likely to do what you want to do. You have at least several decent options:

1) Force film and deinterlace any remaining frames with combing.
2) Honor pulldown flags, do an IVTC in Avisynth and get interlaced video sections down to 23.976p with blend decimation.
3) Make a VFR file (biggest pain in the butt).

[m3mbran3]

thanks for the help Inspector. I'm guessing that by saving my .d2v file with honor pulldown and then using "tfm(order=1).tdecimate()" in my .avs file is what you suggest behind door number 2.

For purely interlaced content I deinterlace with "TDeint(edeint=nnedi2())", if I go for option 1 will this suffice? I am just a bit confused about how it will know which frames to deinterlace.

thanks.

[m3mbran3]

So I ran some tests on one of the troublesome segments, results below.

http://img197.imageshack.us/img197/7517/fgorig.png
original

http://img195.imageshack.us/img195/8333/fg24nnedi.png
forced film, "nnedi2(qual=3)"

http://img198.imageshack.us/img198/3729/fg29tfm.png
29.97fps, "tfm(order=1).tdecimate()"

I know the upload quality isn't great but it appears as if the nnedi2 treated copy is a bit sharper, but I still need to look through the rest of film to see if it has any detremental effects elsewhere.

[Mug Funky]

watch for speed-changes...

remember that this show is a TV series, not a feature film. the main difference is that in a TV series, care needn't be taken to keep things at a consistent 24fps timebase.

also means there's likely to be speed changes, fit-to-fill edits and all manner of pan and scan effects, all likely to be at 60i. if the editor of the show is sane, they'll keep that crap to a bare minimum, but you'll still be seeing stuff like that happening simply to cover errors made during shooting (like zooming a shot in and racking down a bit to cover a wayward boom mic - this happens all the time).

so you'll have to find a way to decimate this without spoiling movement or exacerbating the artefacts that are probably already in the source. it'll also have to gracefully handle orphaned fields and 3:2 cadence breaks that'll be all over this.

btw, the colour looks wrong in that screenshot. maybe check you're in 601 and not 709 space? or maybe the show just has perverse colours (i don't like that look at all, but i guess it's part of the show and shouldn't be altered).

[m3mbran3]

The colour is fine in the actual video, I think it is my process of taking a print screen, pasting in photoshop and uploading to an image host that is making the colours look funky.

I'm currently processing the first episode, firstly with nnedi2 and then with tivtc and am going to compare the output. But from my tests so far both seem to provide pretty similar output. If I use force film then my output file only has ~5% of its frames interlaced according to DGIndex, so mostly the episode looks fine. It probably doesn't help that this is the pilot episode as well so there are probably plenty of corners cut in the editing process.

[Inspector.Gadget]

If you use nnedi2, aren't you just making full frames out of half-height ones? You don't need to deinterlace every frame after force film! You can run TFM against the script once to find out what frames it thinks are combed and then process them.

[m3mbran3]

Yes you are right, don't know how I managed to overlook that. If I run TFM with pp=1 and display=true then I am notified when it detects a combed framed. Is it then possible to use this information to only use NNEDI2 on these frames?

I tried tfm(order=1,display=true,pp=1).nnedi2() but that didn't work.

edit: figured it out with conditionalfilter(last,nnedi2(),last,"IsCombedTIVTC","=","true") although from comparing the output it appears as if NNEDI2 was only decombing those fields that needed it anyway. Anyway after all that I think I will just stick with a 29.97fps .d2v stream and then tfm.tdecimate it.