Questions on IVTC for attached clip with ffmpeg

[aramkolt]

Hello all, I'm working with this clip that I am pretty sure uses some form of inverse telecine.

What ffmpeg script should be used to deinterlace it?

Is my understanding correct that if a clip does use inverse telecine, then other deinterlacing algorithms such as QTGMC should not be used?

Further, if the eventual destination is YouTube and IVTC makes the frame rate something around 24FPS, then how does YouTube handle that content since most YouTube content is either 30/60FPS?

[Sharc]

The clip is 2:3 telecined film. For IVTC to 23.976 progressive film frames using ffmpeg:

ffmpeg.exe -i "ivtc.mov" -c:a aac -c:v libx264 -preset slow -crf 18 -pix_fmt yuv420p -vf fps=30000/1001,fieldmatch=order=tff:mode=pc_n:combmatch=full:cthresh=10,yadif=deint=interlaced,decimate "out_IVTCed.mp4"

Details see
https://ffmpeg.org/ffmpeg-filters.html#fieldmatch

And yes, telecined stuff should not be deinterlaced, but inverse telecined (IVTC).

[davexnet]

This clip looks like it is telecined (film rate > NTSC rate), IVTC undoes it and extracts the clean, 24 fps, frames.
Youtube has no problems wth 24 fps

[aramkolt]

The clip is 2:3 telecined film. For IVTC to 23.976 progressive film frames using ffmpeg:

ffmpeg.exe -i "ivtc.mov" -c:a aac -c:v libx264 -preset slow -crf 18 -pix_fmt yuv420p -vf fps=30000/1001,fieldmatch=order=tff:mode=pc_n:combmatch=full:cthresh=10,yadif=deint=interlaced,decimate "out_IVTCed.mp4"

Details see
https://ffmpeg.org/ffmpeg-filters.html#fieldmatch

And yes, telecined stuff should not be deinterlaced, but inverse telecined (IVTC).

If I wanted to inverse telecine and then keep it as ProRes422HQ (essentially to avoid any compression and make it more appropriate for nonlinear editors), how would the script change?

Thanks everyone!

[Sharc]

The clip is 2:3 telecined film. For IVTC to 23.976 progressive film frames using ffmpeg:

ffmpeg.exe -i "ivtc.mov" -c:a aac -c:v libx264 -preset slow -crf 18 -pix_fmt yuv420p -vf fps=30000/1001,fieldmatch=order=tff:mode=pc_n:combmatch=full:cthresh=10,yadif=deint=interlaced,decimate "out_IVTCed.mp4"

Details see
https://ffmpeg.org/ffmpeg-filters.html#fieldmatch

And yes, telecined stuff should not be deinterlaced, but inverse telecined (IVTC).

If I wanted to inverse telecine and then keep it as ProRes422HQ (essentially to avoid any compression and make it more appropriate for nonlinear editors), how would the script change?

Change the codec and format, probably something like (I didn't try):

ffmpeg -i "ivtc.mov" -c:a aac -c:v prores_ks -profile:v 3 -vendor apl0 -bits_per_mb 8000 -pix_fmt yuv422p10le -vf fps=30000/1001,fieldmatch=order=tff:mode=pc_n:combmatch=full:cthresh=10,yadif=deint=interlaced,decimate "proRes422_output.mov"

[aramkolt]

Thanks, that seems to have worked! I did notice on the ProRes output doing it as above makes the bitrate much higher (close to 80Mb/s) in the resulting video file which is interesting since there are fewer frames per second. Making it output ProRes422 (using -profile:v 2) is much closer to the original HQ bitrate which was about 59Mb/s (comes out to 63.25Mb/s using the ProRes422 -profile:v 2 option).

So in a sort of related question, why would a 59Mb/s file be considered "ProRes422HQ" when it's 29.97FPS but then a 63.25Mb/s file is considered plain "ProRes422" (not HQ) when the frame rate is reduced to 23.97? In my mind, fewer frames per second means even more bitrate is dedicated to each frame?

You'd think the Mb/s would be the determining factor on whether something is HQ vs regular?

[poisondeathray]

You're not comparing the same encoder or settings

Different prores encoder (prores_ks) with different settings (-bits_per_mb 8000) was used and will yield higher bitrate than -c:v prores (the latter is the default ffmpeg prores encoder, which you used for the sample).

All those will be slightly different than other implementations such as official certified Apple Prores . In general, the ffmpeg implementations yield slightly higher bitrates

[aramkolt]

Ah, thanks for that info.

Also, when I actually run the deinterlace script, I get a lot of "Frame is still interlaced" messages. Is that normal? Guessing it might be on the frames that are basically progressive segmented frames (ie no motion between two fields on the same frame, which then basically gives a progressive frame)?

[poisondeathray]

It might "normal" ; it depends on the source

This means is after the fieldmatch filter, ffmpeg is still detecting combing, and in that command above yadif is used to deinterlace only those indicated frames

A perfect telecine on a clean source should have no combing after field matching. You should get zero warnings

But you often don't have "perfect" with broadcast - things are edited breaking cadences (e.g. commercial edits) , noise /compression with interlaced encoding can cause artifacts detected as combing but they might not be (false positives). Too many false positives is not ideal , because that means ffmpeg will be deinterlacing more frames, thus degrading the quality of those frames.

Also other semi-common types of things can cause false positives, such as horizontal patterns - eg. window blinds. If those get detected you will get aliasing because of the deinterlacing applied

Sometimes you might have to adjust the threshold for combing detection. If the threshold is too high, you might miss a combed frame and that looks bad too.

In your sample, the cadence was broken from the way it was cut. It starts with a combed frame, then 3 clean, then 2 combed and so forth. The 3:2 is normal, but the 1st frame was cut incorrectly interupting the cadence. There is a single field at the beginning that is missing it's field pair. ie. it's an "orphaned field". You would expect to get a "still interlaced" warning at that timestamp