Best frame rate upscaler?

[Mephesto]

A few years ago I asked and got SmoothFPS which was good enough for the video I was working with at the time. I'm wondering if there have been advances since then. Is SmoothFPS still the best out there?

MSU has very promising FPS upscalers but they are unavailable for individual use. Those bastards.

[poisondeathray]

All the avisynth based ones are similar IMO (they tend to fail with the same artifacts on the same types of conditions), but you can try svpflow or interframe which are "supposedly" slightly better . svpflow has benefit of GPU acceleration

http://www.svp-team.com/wiki/Plugins:_SVPflow
http://forum.doom9.org/archive/index.php/t-164554.html
http://forum.doom9.org/showthread.php?t=160226

The only way I've seen around the bad morphing artifacts is masking, guides and foreground/background mattes in something like twixtor pro (and it's a lot of manual tweaking) - arguably this is the "best" way

[Mephesto]

Thanks PDR. This one is just as amazing now as SmoothFPS was 3 years ago. I've tried it to view movies I had in real-time and the artifacts are only noticeable upon inspection, but not playback. It looks smooth and very genuine.

However, I just needed it to convert a 20 fps video to 30 fps. I don't really care for 60fps video. Call me ******* crazy but I prefer the cinema 24/30fps. 60 fps makes the video look synthetic and CGI-ish. I like 60 fps on video games and cartoons, but not film.

Maybe I'm a masochist who got used to the low FPS cinema uses today, who knows.

Anyway, is there a way to customize the framerate with InterFrame? Right now I have to do AssumeFPS tweaks so it won't convert my 20 fps to 60 fps, which isn't hard to do but it makes me have to do math with windows calculator.

EDIT: Just an observation, the author needs to implement bi-directional frame prediction to avoid artifacts on the left and right edges of the video during scrolling.

[lordsmurf]

Hardware is still best.
This is one of those things where software still fails.

[Mephesto]

Hardware is still best.
This is one of those things where software still fails.

Why is that? They're both making something out of nothing, no?

[Mephesto]

Btw, can this upscaling algorithm be used to replace corrupted frames? The one in this thread is horrible. https://forum.videohelp.com/threads/344962-VHS-Cleanup-What-can-i-do?p=2152073&viewfull=1#post2152073

[manono]

The more consecutive frames being created and the more complex the frames (the more movement) the more likely it is to fail. I find that with a single frame and not too much movement, it's almost perfect. I use Gavino's for single frames and the one jagabo points out in the link for consecutive bad frame replacement. Here's Gavino's FixBadFrames function:

function FixBadFrames(clip c, string frames) {
# Replace each frame from a list of 'bad' frames by using MFlowInter to interpolate
# between the nearest pair of 'good' frames
  c
  sup = MSuper()
  bv = MAnalyse(sup, isb=true, delta=2)
  fv = MAnalyse(sup, isb=false, delta=2)
  candidates = MFlowInter(sup, bv, fv, time=50.0, ml=100).DuplicateFrame(0)
  ReplaceFramesSimple(candidates, mappings=frames)
}

#Usage FixBadFrames("34 64 96")

It also needs stickboy's RemapFrames.dll:

http://avisynth.org/stickboy/

[Mephesto]

The one I found in the first link was really horrible, especially on consecutive frames, I'd rather just delete the bad frames than mess with it at all. What I was asking is if there was a way to replace bad frames using the same algorithm that advanced upscaler PDR linked to uses.

[poisondeathray]

There is no wrapper function for svpflow for frame replacement, but you can just combine it with clipclop or remapframes

As mentioned earlier, they tend to fail in the same circumstances. You can adjust the settings or do iterations with different settings and combine/composite them

[Mephesto]

Oh I forgot to say, I used the third link you provided, that InterFrame. The interface for SVP confused me.

Is InterFrame based on SVP?

[poisondeathray]

Is InterFrame based on SVP?

No, but all of them are based on mvtools2