Log in

View Full Version : Resizing hybrid clips


Aegwyn11
15th May 2013, 05:48
I have a very hybrid clip that is composed of a combination of film and video content. The clip is encoded as 1080i29.97, with the video content interlaced normally and the film content telecined. I need to resize the clip to 720p and 480i. My final output needs to be CFR, just like the input clip, and it can take however much time is needed to process. What is a good way to handle the needed combination of deinterlacing and inverse telecining?

kolak
15th May 2013, 15:14
Deinterlace with QTGMC to double fps, scale and intetlaced back when needed.

Aegwyn11
15th May 2013, 16:40
Deinterlace with QTGMC to double fps, scale and intetlaced back when needed.

My fear with this process is that it introduces loss. For the film sequences, I don't need to interpolate missing fields as the original progressive frames are there, just telecined. Plus, I'm not sure if QTGMC would screw up the film sequences since they aren't really "interlaced".

My idea process would be something smart that could let me deinterlace the video sequences and inverse telecine the film sequences (resulting in something VFR), do the resizing, then re-telecine the film sequences and (if necessary) re-interlace the video sequences, resulting in a final clip that is CFR. But short of sitting down and manually chopping my sequence into individual clips (which would take FOREVER), I have no clue how to do this or if its even possible using currently available tools.

kolak
15th May 2013, 17:22
Yes, it will affect them, but with good settings loss is not that big.
Another idea is to use this script:


clip2=QTGMC(preset="very fast", NoiseProcess=1, NoiseRestore=0.8, Sigma=2.0,dct=3, sharpness=0.1).selecteven()
TFM(mode=0, PP=2, cthresh=2, mi=60, mChroma=false, blockx=16, blocky=16, clip2=clip2)

This will detect progressive frames and bypass them- for the rest of the source it will use QTGMC for deinterlacing. Problem is that this will deinterlace to single fps- not sure how to change TFM part, so it will do double fps. By adjusting cthresh and mi you can get 95% accuracy with detection of progressive parts. You should make sure all is deinterlaced in cost of some progressive part being also processed (not big quality loss) than opposite.