View Single Post
Old 6th June 2007, 14:30   #1068  |  Link
plugh
A hollow voice says
 
Join Date: Sep 2006
Posts: 269
Quote:
the method you came up with is probably the simplest. There isn't an easy way to modify tdecimate to do this internally.
OK, thanks.

The mode4 / mode2 combo with the de-jittered stream has produced the best decimation results so far, but it still isn't quite right. I *have* identified the telecining pattern used in this 720p broadcast - 576 film frames into 1409 video frames, which works out to 59.94*(576/1409) = 24.503506032, which yeilds
Quote:
rate = 24.503506 actual rate = 24.503515
mode2_num = 1 mode2_den = 2 numCycles = -20 clength = 12
mode2_cfs 0 = 12
mode2_cfs 1 = 132
mode2_cfs 2 = 3564
mode2_cfs 3 = 99792
or (60000/1001)*(576/1409) = 24.503530536, which yeilds
Quote:
TDecimate: mode 2 error, number of frames after decimation doesn't match!
Specifically - the dup counts are
Code:
212121212121212121 45->18
112121212121212121 44->18
112121212121212121 44->18
202121212121212121 44->18
202121212121212121 44->18
211121212121212121 44->18
211121212121212121 44->18
212021212121212121 44->18
212111212121212121 44->18
212111212121212121 44->18
212120212121212121 44->18
212121112121212121 44->18
212121112121212121 44->18
212121112121212121 44->18
212121202121212121 44->18
212121211121212121 44->18
212121211121212121 44->18
212121212021212121 44->18
212121212111212121 44->18
212121212111212121 44->18
212121212120212121 44->18
212121212120212121 44->18
212121212121112121 44->18
212121212121112121 44->18
212121212121202121 44->18
212121212121211121 44->18
212121212121211121 44->18
212121212121212021 44->18
212121212121212021 44->18
212121212121212111 44->18
212121212121212111 44->18
212121212121212120 44->18
The most notable 'feature' of the resulting decimation (using 24.503506032) is that sometimes tdecimate eliminates three frames in a row (you can see runs of three telecine dups never occur in the above pattern) and thus ends up including dups elsewhere (see attached images - you can clearly see where dups are getting through). Thus, screen motion is jerky in places.

So my question is, how can I get this 'right', short of hand crafting an ovr file? Are the decimation cycles being derived by mode2 not quite correct? Is there any way to tell it 'max dup length=2' (if that would help)? Or am I doing something fundamentally wrong

Thanks in advance
Attached Images
  
plugh is offline   Reply With Quote