Deinterlace on/off automatically on a continuous stream?

[ctcwired]

Got a continuous SDI stream (Decklink input), in this case a program feed at a TV station I am live streaming, which I want to always output progressive. Some live content is genuine 1080i29.97, while other content is 1080PsF (1080p29.97 but being played out through a 1080i router).

BWDIF does a great job on the real interlaced content, but I find it tends to add artifacts to PsF stuff, particularly on edges of logos and text if they're animated.

Is it possible to chain together some detection to have a sort of rolling average of confidence on whether the content is interlaced or not, and have BWDIF activate & deactivate automatically?

I'm thinking something like:

-filter:v "idet=half_life=300,bwdif=mode=0,deint=interlaced"

But I'm not sure if the metadata or filter behavior actually works this way?

In my testing with idet it does seem to log "tff" (top field first) and "progressive" fairly reliably on the multi-frame report, but less so on the single frame. So I'm not sure if I can use the "half_life" parameter in this way. I've also seen references to using the "fieldmatch" filter for this type of thing, but that seems more for inverse telecine purposes so I'm not sure.

Thanks

[hello_hello]

I don't use ffmpeg for that sort of thing but I'd give the fieldmatch filter a try.

Field matchers have to repair any residual combing that remains after field matching by deinterlacing. For the fieldmatch filter, if some of the frames are interlaced or contain combing after field matching, it should mark them as interlaced, and according to the docs, only the frames marked as such are de-interlaced, so the progressive frame should remain unmolested. That's the way I read it.....

https://ffmpeg.org/ffmpeg-all.html#fieldmatch
https://ffmpeg.org/ffmpeg-all.html#Deinterlacing

[ctcwired]

That's the way I read it as well. I've also since learned of the "kerndeint" deinterlacer, which has worse performance but is more flexible configuration for motion detection thresholds. I'll give them both a try later today. Thanks for that suggestion.

Of course the most appropriate option would be to pick up a Teranex, but that's a management decision.