[MEncoder-users] Newbie struggling with pullup

Scott Larson scowl at pacifier.com
Thu Feb 28 06:19:47 CET 2008


On Wed, 2008-02-27 at 22:15 -0600, Laine Lee wrote:
> But in 30000/1001 output, I always see a pattern of 5 frames with the 4th
> and 5th being similar (duplicate). Would decimate have a better chance of
> giving me output with only unique frames?

The common problem with using decimate with 720p is that you can't
always count on the duplicated frames being true duplicates or even
close duplicates. If you step through certain scenes and follow the 3:2
pattern, you'll often find that frames that should be duplicated aren't
quite duplicates of the previous frames, as if the station's encoder is
adding detail to the previous frame in an attempt to "catch up". Many of
these frames that should be duplicates are often more different to
decimate than frames that *are* truly different frames.

This kind of frame is most common in scenes with a lot of action. Even a
change of scene will sometimes generate an ugly frame followed by a
perfect one. I'm guessing this is because the encoder is sending
fixed-length GOPs and the change of scene is falling in the middle of
one. Or perhaps I shouldn't speculate after the intra/inter/keyframe
discussion from yesterday. I admit I don't have a clue what the real
cause is.

Unfortunately decimate doesn't know which frame is the "better" frame.
It only knows they're different so it will pass them both. If this
happens you'll get frames which will appear to be duplicates to your
eyes and you'll see stuttering. You also can't rely on -ofps to always
pick the right ones because if it's one frame off of the 3:2 pattern, it
will pick the first and last frame in the triple duplicates and you'll
have terrible stuttering.

Using tinterlace=4 with pullup has been the most reliable method for me.
Sometimes I'll see what looks like interlacing in a frame. I think is
caused by an ugly frame being combined with a better frame. It's only
one frame and is not too distracting.




More information about the MEncoder-users mailing list