Quote:
Originally posted by tritical
Nope, I wouldn't mind... you can add whatever you feel like  .
|
Ugh. Well, I would have already done something if I hadn't underestimated the time marking 24 MicroJava-compilers for my part-time tutor job at the university takes...
I'll try to start on it this weekend even though I have to come into work on Saturday.
Anyhow, I just tried TIVTC on the first episode of Ghost In The Shell: Stand Alone Complex and found that there was a duplicate frame causing a stutter after a few scenechanges, just as it happened with Decomb on other sources.
I'm wondering - since you're already doing scene change detection for the video parts, would it be possible to detect them for the film parts as well and treat them specially?
I was thinking about something like this:
If you find a scenechange and on one side of it there are more than <cycle> dups in a row, look for a frame to drop on the opposite side of the scenechange first and drop a frame there if you can find one - this would take care of scenechanges that go from a still scene to a moving one.
On the other hand, if the scenechange is badly edited so you'd have to drop a duplicate on both sides (which might be detectable by looking ahead and back a bit and finding out the used pattern, or just by checking if there's 2 or more candidates within <cycle> frames of the scenechange in both directions) you could drop the one closest to the scenechange on each side and insert a new frame at the scenechange's position that blends the 2 neighbouring frames so audio & video stay in sync. IMHO that would be much less noticeable than a sudden jerk in a pan that just started or ended.
I tried cutting out a scene or two to illustrate this, but everytime I cut out a part with Cuttermaran, run it through DGIndex and load it in an AviSynth script, I get different metrics in TDecimate due to different combing...
(and yes, I did cut at an I-frame and chose starting points that didn't cause a shift of the cycles that TFM gets...)
Well, this is just a suggestion; I'll see if I can find a way to cut out parts reliably (help, anyone?) and upload you an example or two if I manage to do it - if you want, of course.
np: Autechre - Latent Quarter (Envane)