Alternatives to MotionProtectedFPS and MVFlowFPS?

[Chainmax]

I tried both these filters for a 15fps-->29.97 fps conversion, but the former gave me unsatisfactory results and the latter gave me access violation errors after many tries. What else could I try that would give me good results?

[AI]

framedbl (http://www.avisynth.org/warpenterprises/files/framedbl_25_dll_20030621.zip)

[MrTroy]

Yes, doubling the frames and then AssumeFPS(29.97,sync_audio=true) seems the best solution.

[Chainmax]

Wouldn't it be better to use bobbing instead of frame doubling?

[MrTroy]

Wouldn't it be better to use bobbing instead of frame doubling?Well, since 29.97fps is almost exactly twice 15fps, doubling the framerate should provide a perfectly smooth result. Not smoother than the original file of course, you can't create fluidity out of nothing.

[Alain2]

Not smoother than the original file of course, you can't create fluidity out of nothing.
That's where motion compensation is supposed to step in ;)

[Chainmax]

Are there any motion compensated framerate changers besides the ones I mentioned?

[foxyshadis]

Twixtor for Premiere (and After Effects, Final Cut Pro, Shake, and Combustion), and expensive hardware changers that suck even more. But for avisynth, Motion and MVFlow are the only choices.

Twixtor 4.5 is badass, way better than our two, but expensive and a little slower. (I only used the trial.)

[Chainmax]

Crap, I wish I could find out why MVFlowFPS gives me access violation errors :(.

[actionman133]

What is the resolution of the clip your testing MVFlowFPS with? I've found that I got some unusual behavior if the width and height weren't conforming to something like mod 8 or mod 16 values...

[communist]

Wouldn't it be better to use bobbing instead of frame doubling?
Bobbing would only work (=make sense) if its interlaced.
Latest MVTools is pretty good - so if you can get it to work that would be a good start :)

[morsa]

Some time ago I experimented using MotionProtectedFPS, MVflowFPS, Dynapel's MotionPerfect and TooT.
Let me teel you that the results were excellent 80% of the time.
What I noticed is that if we still had the option of using the old block system old MVtools had, that 20% would be much lower.
So, Fizick, if you are reading this, please take note ;)

[videoFred]

Some time ago I experimented using MotionProtectedFPS, MVflowFPS, Dynapel's MotionPerfect

Yep, I did the same.
MVflowFPS wins, if you ask me.
I use it all the time.

Of cource you see the artefacts on computer, if you analyse the frames, but you barely see them on TV.

But maybe we could even fine tune it...
Now, what would be better.. denoising before or after MVAnalyse..
Stabilising with Depan before or after...or both?
What's the effect of stabilising on MVAnalyse?
Is motion estimation better, then?
Stand alone Depan or MVDepan?
And what are the best settings for optimal motion estimation etc... etc...

Fred.

[foxyshadis]

You should analyse a denoised (but not de-detailed) version and apply the results to the original, if at all possible. Beyond that I'm not terribly familiar with mvflow's idiosyncrasies.

I've seen some very interesting results from combining mesh-warping and block-matching, it may have been a didee script or a pdf, I don't remember.

[videoFred]

You should analyse a denoised (but not de-detailed) version and apply the results to the original, if at all possible.

This is perfect possible.. But then my original is still not denoised... Then denoise again the result? This is going to slow down the script...

Fred.

[Didée]

What I noticed is that if we still had the option of using the old block system old MVtools had, that 20% would be much lower.
So, Fizick, if you are reading this, please take note ;)
If MVTools had an option to use calculated motion vectors in inverse direction, probably I could put something together. The MC denoiser on my workbench uses a method that produces a mostly sufficient reliable error mask of those areas where the compensation [frame_N]-->[vector_N_NN]-->[frame_NN] outputs a "unusable" result. However, since the error mask is just a spatial mask covering [frame_NN], it would be needed to make a compensation of the error mask in reverse direction: [frame_N]<--[vector_N_NN]<--[error_NN].
With the filters currently in MVTools, this is not possible to do.

And yes, with all the focussing and praising of the new "flow" compensation methods, the old blockbased methods IMO have been neclegted too much. Block-based and warp-based methods both have their pros and their cons. As of now, I don't see the warping method in a position being "generally superior". Both methods do have their sweetspots, they're just in different places.

[Mug Funky]

@ Didée:

block based seems to perform better with occlusion. i wonder if mvflow could be modded to fall back on block moving when "tearing" is detected?

"flow based" MC works wonderfully on large rotating objects :)

but you're right about them both having their merits. for example, removing spots in film seems better done with mvcompensate, but framerate change looks better with mvflow (except with occluding objects). paradox :)

i'd love for something fast to come up though - motionprotectedfps is fast enough for now, but doesn't like scenechanges, so i tend to only use it on talking heads or pure video with no cuts.

perhaps i should start saving up for a Snell and Wilcox Alchemist PhC? or start being reeeally nice to my boss?

[Fizick]

Of course, block based compensation may be improved, with right block overlap processing.

[Revgen]

i'd love for something fast to come up though - motionprotectedfps is fast enough for now, but doesn't like scenechanges, so i tend to only use it on talking heads or pure video with no cuts.


You mean that the finished video doesn't look fast?

Or do you mean that it's slow to process?

MVFlowFPS2 is fast enough for me.

[morsa]

Well, as long as I have the option of three different types of frame interpolation I can surely use TooT.DLL (Two Out Of Three) filter to get a corrected image.

[Chainmax]

How would you compare DePan with the ones mentioned so far?

[Clown shoes]

http://www.snellwilcox.com/news_events/press_releases/183

According to the guy at the Snell and Wilcox booth at Broadcast live, Helios is a software version of Alchemist, using thier patented Ph.C phase correlation motion estimation system. I think he said it was going to retail for about £3,000.00 but compared to the alchemist's price tag in excess of £100,000.00, it seems like a steal.

[Chainmax]

Lol, I could buy two very nice computers for that price :). In any case, thanks to the help of scharfis_brain I created a function that works wonderfully and shows no artifacts at all. Thanks anyway :).

[Floatingshed]

Lol, I could buy two very nice computers for that price :). In any case, thanks to the help of scharfis_brain I created a function that works wonderfully and shows no artifacts at all. Thanks anyway :).

What is that function? And can I have a play with it please?
Thanks.

[shoopdabloop]

three years. snazzy.

and i too would like to see this function, if it still works. :D

[Alex-Kid]

What about motion blur as an acceptable solution? I think there is too little temporal information for smooth frame interpolation.

[jeffy]

What is that function? And can I have a play with it please?
Thanks.
http://forum.doom9.org/showthread.php?p=850027#post850027
and here
http://forum.doom9.org/showthread.php?p=850095#post850095