Thread: VirtualDub2
View Single Post
Old 12th December 2020, 20:28   #1079  |  Link
poisondeathray
Registered User
 
Join Date: Sep 2007
Posts: 5,346
Quote:
Originally Posted by nji View Post

But otherwise (as with DeShaker) there is a problem:
Using the non-linear timeline as linear will (probably?)
lead to bad deshaking results.
So no way that way.
Converting the vfr to cfr will lead to duplicates
(which are the real life timeline anyhow).
But Deshaker produces wrong results at still scenarios.
Which is a pure DeShaker problem
that has nothing to do with vfr, but would also occur at cfr.

Am I right with that?

It depends on your video, what type of VFR , and the reason for the VFR.

VFR to CFR with duplicates is not the real, real timeline (The original, original CFR data before your video). You have missing data. Inserted duplicates are a method of approximation for the real data

In general, the motion will be smoother if you interpolate the missing frames using optical flow (but predisposed to artifacts) - it sounds like you discarded that.


Deshaking either VFR as is ; or converted to CFR (with inserted duplicates) can both produce bad results . It partially depends on how variable your video is (how bad the drops are, the magnitude of the fps drops).

Some types proper timecode VFR video can be the result of low motion in the first place. E.g. still scene. Or you do a camera pan and slow down to a still shot. It's properly represented by fast motion, to slower motion, then no motion.

Other types of timecode VFR are functional - the recording system can't keep up and frames are dropped. It might not be an actual "still scenario" (there is supposed to be motion) , but inserted duplicates represent it as a "still scenario", even though it is not actually. The compensatory motion introduced by a stabilizer can sometimes look wrong
poisondeathray is offline   Reply With Quote