Frame-rate converter that does not disturb cuts or fades and lighting point

[VideoPlayout11]

I am needing to convert some PAL 25fps clips to 50fps and 24fps to 50fps all SD video. I have tried Twixtor by RE:vision but has really messes any cut, fades and lighting changes, you would have to manually edit each point, and that is really time consuming considering my clips are 50mins long. So is there any good professional frame-rate converter MAC based that can assist me. So far I found no luck in finding any .

Thank you in advanced.

[Baldrick]

I'm moving you to our mac section.

[houdini2]

JES Deinterlacer.

houdini

[jagabo]

really messes any cut, fades and lighting changes

That sounds more like a bitrate issue than a frame rate conversion issue. Maybe you should upload a sample that shows the problem.

[VideoPlayout11]

This is an example shows a video converted from After Effects, from 25fps to 50fps using frame blend. See how there is "gooey" motion, that is my issue. Along with cuts and fades and lighting changes. I need a frame rate converter that does not do this, but still get smooth results and no lose of quality in motion

Example of a Line up graphic (mov) converted to 50fps;

[poisondeathray]

For 25p to 50p you only have 3 basic options or approaches

1) frame duplicates
Retains the same motion characteristics as 25p . It is 25p , with twice the # frames, but twice fps

2) frame blends
Inserted frames are blurry, usually a 50/50 mix of before and after

3) optical flow e.g. twixtor, pixel motion in AE, kronos
Attempts to generate a new "inbetween" frame . So the motion will be the smoothest, almost like native 50p . The problem is edge morphing artifacts were the motion vector prediction fails . Sometimes it works great, other times it fails miserably - it's really content dependent

Long story short, there is no good way. It's all pros and cons. Frame duplicates are the "safest" , but the motion is the choppiest . Optical flow is the "smoothest", but the slowest to process (CPU intensive) and can be riddled with artifacts . Method #1 won't mess with edits or fades provided you do it correctly



24p to 50p is a completely different ballgame, because 24 is not evenly divisible into 50 . So you will get an irregular cadence. I would actually speed it up to 25p (both video & audio) then decide on method 1,2, or 3

This is an example shows a video converted from After Effects, from 25fps to 50fps using frame blend.

That example you uploaded didn't use "frame blend" , it's using method #3 optical flow (you're probably using pixel motion or timewarp in AE or twixtor)

[VideoPlayout11]

There is a Frame converter called "ISOVideo" that does a great job converting any frame rate, with smooth results without messing cuts etc: http://www.isovideo.com/legato.php

Only issue its not MAC based. I would have liked to find a software close to this for mac.




This is the setting used in AE when using 'Frame-Blending'

[Herve]

just a question: what for?

I am needing to convert some PAL 25fps clips to 50fps

25->50 = duplicate each frame

and 24fps to 50fps

24->50 = buggy duplication (it is less bad to duplicate each frame = 48fps and to strectch/expand its duration to fit to 50fps)

so, (re) my question: what for? (which use, did you hope frames blending during conversion?, etc)

bye

[jagabo]

That example you uploaded didn't use "frame blend" , it's using method #3 optical flow

I agree.

There is a Frame converter called "ISOVideo" that does a great job converting any frame rate

I'm sure if you ran it on the same source video the end result would look similar.

[VideoPlayout11]

I'm sure if you ran it on the same source video the end result would look similar.

No, if you download one of the examples used, they have converted a 30fps clip, to 60fps without any issues.


- http://www.isovideo.com/legato.php

[jagabo]

I'm sure if you ran it on the same source video the end result would look similar.

No, if you download one of the examples used, they have converted a 30fps clip, to 60fps without any issues.

- http://www.isovideo.com/legato.php

No, they cherry picked their examples. They are not working with your source.

[poisondeathray]

This is the setting used in AE when using 'Frame-Blending'

That's "pixel motion" . Method #3 . When there is a dotted line slanted the opposite way, it's frame blending (50/50) split (Method #2) . When you have frame blending off for the comp (when you see the larger button turned off), then it uses frame duplication (Method #1)

The artifacts you're referring to in your sample are "edge morphing artifacts" from optical flow

[poisondeathray]

I'm sure if you ran it on the same source video the end result would look similar.

No, if you download one of the examples used, they have converted a 30fps clip, to 60fps without any issues.

- http://www.isovideo.com/legato.php

No, they cherry picked their examples. They are not working with your source.

Exactly . They will have similar artifacts depending on type of content

When objects cross over each other, there is no way the software can tell what is what. Motion vectors prediction fails and you get those edge occlusion artifacts . You might have slightly better or slightly worse artifacts with different software, maybe slightly better scene change detection - but you WILL get artifacts. Even custom in house VFX solutions that cost many times more will have these types of artifacts . Manual rotowork and matte guides are required to make them "perfect" . There is no 100% perfect automated solution

Here is one of their examples look at the frame before and after. The software gets "confused" by the trees because the software cannot determine accurately the edges of the trees vs. the people

[jagabo]

Bad-case examples of motion interpolation with other software:

https://forum.videohelp.com/threads/352741-Frame-interpolation?p=2215502&viewfull=1#post2215502

The same software delivers quality much like isovideo's with other videos.

https://forum.videohelp.com/threads/341052-Juddering-in-encoded-HD-video?p=2126815&view...=1#post2126815

[VideoPlayout11]

Thank you all for your help. Looks like there is basically no solution for "smooth results" without any issues. Do you think in future reference, they can make a software capable of pixel motion without distortion?

[poisondeathray]

Thank you all for your help. Looks like there is basically no solution for "smooth results" without any issues. Do you think in future reference, they can make a software capable of pixel motion without distortion?

No . It's just not possible for an automated solution . There will always be situations where the motion vector prediction fails .

If there was "perfect" motion interpolation available, there would be no reason to shoot fast frame rates. No reason to spend $100K on slow motion FX cameras

Maybe in 200 years when we can teleport to other planets and travel faster than light.

[isovideo]

Thank you poisondeathray for pointing out the blemish, but we have made significant improvements since the year-old example from isovideo.com site.

[poisondeathray]

@isovideo - That's good. Maybe it would be a good idea to update your "examples"

Other optical flow methods do not "fail" as poorly on that old frame for parkrun either (e.g. mvtools2, twixtor, kronos)

But all algorithms will have some "blemishes" or motion vector prediction fails in other areas . There will be no perfect automated solution

[isovideo]

The examples have been updated! While we agree that all motion estimation systems can have issues depending on image content, it should be pointed out that all motion algorithms must strike a balance between quality and speed. The tricky part is designing a quality algorithm that can be efficiently mapped onto GPUs. The above results were obtained running on a single workstation in realtime, whereas all the other technologies you mention are VERY slow. Furthermore, I doubt any of these algorithms could be ported efficiently onto GPUs.

[poisondeathray]

While we agree that all motion estimation systems can have issues depending on image content, it should be pointed out that all motion algorithms must strike a balance between quality and speed. The tricky part is designing a quality algorithm that can be efficiently mapped onto GPUs. The above results were obtained running on a single workstation in realtime, whereas all the other technologies you mention are VERY slow. Furthermore, I doubt any of these algorithms could be ported efficiently onto GPUs.

Yes for most users the speed is important. , and ME algorithms can be very slow

Kronos runs on Cuda (Nvidia) faster than real time on HD content on a single socket workstation, but is prone to the same edge morphing artifacts that require user intervention and mattes to fix . Twixtor runs with GPU assist as well (but it's not as fast)

The OP wanted some automated "perfect" solution which doesn't exist (and probably never will)