Why don't my interpolated videos look as good as examples I see on youtube?

[jagabo]

If you can't get SVP to work to your liking why don't you just try one of the alternatives?

[bradwiggo]

If you can't get SVP to work to your liking why don't you just try one of the alternatives?

What are the alternatives? you mean things like framerateconverter?

The reason I am using SVP is the person who made those youtube videos said they did as well.

[jagabo]

Maybe the GPU version of SVP works better than the CPU version. If you have a supported GPU try it.

[bradwiggo]

Maybe the GPU version of SVP works better than the CPU version. If you have a supported GPU try it.

I will see if I can get that to work. Did the issues I mentioned in my other comment make any difference to the smoothness?

Also, looking at this table here: https://www.svp-team.com/wiki/GPU_Compatibility

I assume my GPU is okay, as it says AMD A series IGPs (I assume integrated GPUs), and I have a AMD A10-7300 with an integrated Radeon R6). The only thing is it says use an older driver, so I may need to dig that up out of somewhere (the internet).

[johnmeyer]

I can't believe how long this thread has gone on.

Folks, motion estimation is always going to produce a lot of ugly frames. Don't get me wrong, it has its place, and I use it all the time, but the OP is chasing fairies in this pursuit, by which I mean that this isn't ever going to produce a result that will make him (or other people) happy. This is not his fault, other than thinking that there is a way to improve these videos without causing problems that might be worse than what he is trying to fix.

I've provided lots of help with these pursuits in the past, and my work is referenced in this thread:

Best Motion Interpolation's software/plugin ?

As you will see in that thread, and the threads it references, there are some settings that will minimize the artifacts, but even with the addition of masks and other more complicated attempts to fix the inevitable ME problems, you still get a lot of ugly frames.

You definitely should read those threads because there were lots of ideas given that might help with a particular scene.


My advice: watch and enjoy the original video, and don't spend a second month in pursuit of the unicorn.

[bradwiggo]

I can't believe how long this thread has gone on.

Folks, motion estimation is always going to produce a lot of ugly frames. Don't get me wrong, it has its place, and I use it all the time, but the OP is chasing fairies in this pursuit, by which I mean that this isn't ever going to produce a result that will make him (or other people) happy. This is not his fault, other than thinking that there is a way to improve these videos without causing problems that might be worse than what he is trying to fix.

I've provided lots of help with these pursuits in the past, and my work is referenced in this thread:

Best Motion Interpolation's software/plugin ?

As you will see in that thread, and the threads it references, there are some settings that will minimize the artifacts, but even with the addition of masks and other more complicated attempts to fix the inevitable ME problems, you still get a lot of ugly frames.

You definitely should read those threads because there were lots of ideas given that might help with a particular scene.


My advice: watch and enjoy the original video, and don't spend a second month in pursuit of the unicorn.

I agree that motion interpolation can never be perfect, however, I am currently just trying to get it to look as good as I know it can, as I have seen videos on youtube that are better than anything I have managed to produce. I don't really see it as chasing the impossible, as I know it can be done (just not how).

[manono]

I don't really see it as chasing the impossible, as I know it can be done (just not how).

Sounds like chasing fairies to me. Just because it worked well on one video doesn't mean it'll work well on another. All kinds of things might throw it off - movement too fast, movement of one object in front of another, movement from offscreen to onscreen, all kinds of things. I work with interpolation virtually every day of my life as I work with old films with a ton of artifacts. If there are frames or a series of frames on which it doesn't work well, try using frame replacement rather than frame interpolatoion.

I agree with johnmeyer that this thread has gone on for far too long.

[amaipaipai]

This is a discussion forum, let them discuss as long as they want.
If this is bothering you, unsubscribe and don't read it.

[bradwiggo]

I don't really see it as chasing the impossible, as I know it can be done (just not how).

Sounds like chasing fairies to me. Just because it worked well on one video doesn't mean it'll work well on another. All kinds of things might throw it off - movement too fast, movement of one object in front of another, movement from offscreen to onscreen, all kinds of things. I work with interpolation virtually every day of my life as I work with old films with a ton of artifacts. If there are frames or a series of frames on which it doesn't work well, try using frame replacement rather than frame interpolatoion.

I agree with johnmeyer that this thread has gone on for far too long.

I get it doesn't work as well on some videos, but the video I am comparing it to is the same.

I don't see how the thread has gone on too long, I haven't solved my problem yet, and I was getting there. Does it matter?

[amaipaipai]

I don't see how the thread has gone on too long, I haven't solved my problem yet, and I was getting there. Does it matter?

Don't bother.
The forum rules doesn't specify any rule against the amount of posts a user can or can't make.

Keep posting as long as you want. [emoji1303]

[johnmeyer]

I agree with the others that you should keep on posting, if you want. I was simply trying to say, as nicely as I know how, that you have gotten to the point of diminishing returns, and you are wasting your time.

You say that you "know it can be done." How do you "know" this? My point in posting was not to be snarky, but to point out that, based on that thread I linked to (did you read it??), I am 99% certain that it cannot be done. Dozens of people who are far smarter than I am have tried, and they still haven't come up with a solution that is any better than that series of settings for MVTools2 that I stumbled upon, and which is still used as a benchmark of sorts. However, even those "magic setting" fail quite spectacularly all the time.

The same can be said for the very expensive commercial motion estimation (also called optical flow) tools, such as Twixtor, which cost $330. It is no better than MVTools2 and its SVP and Interframe derivatives.

The problem is that motion estimation cannot deal with things that suddenly appear behind another object; it cannot deal with opposing motion, such as the legs of a person walking across the field of view, fairly close to the camera. It cannot deal with a horizontal pan across a field of objects that have a strong vertical component, such as a picket fence. And it can fail spectacularly with well-defined, but fairly small features that are bent and curved (see the moose antlers in the video I link to below).

You can see all of these failures in this 1940 film I transferred of a parade in Flint, MI. It was taken at 16 fps and, because there is a lot of horizontal panning, that low frame rate produces a lot of visual judder, a visual artifact that is formed in your brain when the fps is below the threshold for human persistence of vision. So, I increased the frame rate to 30 fps using motion interpolation. If you click on this link, you won't see any judder, but you will see every single one of the artifacts I just mentioned. I decided that, for this particular clip, those horrible artifacts were more acceptable than the judder. Most of the time I do not prefer this tradeoff, and so I deliver the film transfer in its native fps.

https://www.youtube.com/watch?v=t8HjRN0rw5M

So keep posting, if you want, but what you are going to find is that if you manage to find a setting that reduces the artifacts with one clip (it almost certainly will not eliminate them), those new settings will then produce artifacts on another clip that your first group of settings dealt with just fine. It is a never-ending game of whack-a-mole.

[bradwiggo]

I agree with the others that you should keep on posting, if you want. I was simply trying to say, as nicely as I know how, that you have gotten to the point of diminishing returns, and you are wasting your time.

You say that you "know it can be done." How do you "know" this? My point in posting was not to be snarky, but to point out that, based on that thread I linked to (did you read it??), I am 99% certain that it cannot be done. Dozens of people who are far smarter than I am have tried, and they still haven't come up with a solution that is any better than that series of settings for MVTools2 that I stumbled upon, and which is still used as a benchmark of sorts. However, even those "magic setting" fail quite spectacularly all the time.

The same can be said for the very expensive commercial motion estimation (also called optical flow) tools, such as Twixtor, which cost $330. It is no better than MVTools2 and its SVP and Interframe derivatives.

The problem is that motion estimation cannot deal with things that suddenly appear behind another object; it cannot deal with opposing motion, such as the legs of a person walking across the field of view, fairly close to the camera. It cannot deal with a horizontal pan across a field of objects that have a strong vertical component, such as a picket fence. And it can fail spectacularly with well-defined, but fairly small features that are bent and curved (see the moose antlers in the video I link to below).

You can see all of these failures in this 1940 film I transferred of a parade in Flint, MI. It was taken at 16 fps and, because there is a lot of horizontal panning, that low frame rate produces a lot of visual judder, a visual artifact that is formed in your brain when the fps is below the threshold for human persistence of vision. So, I increased the frame rate to 30 fps using motion interpolation. If you click on this link, you won't see any judder, but you will see every single one of the artifacts I just mentioned. I decided that, for this particular clip, those horrible artifacts were more acceptable than the judder. Most of the time I do not prefer this tradeoff, and so I deliver the film transfer in its native fps.

https://www.youtube.com/watch?v=t8HjRN0rw5M

So keep posting, if you want, but what you are going to find is that if you manage to find a setting that reduces the artifacts with one clip (it almost certainly will not eliminate them), those new settings will then produce artifacts on another clip that your first group of settings dealt with just fine. It is a never-ending game of whack-a-mole.

It's not really artefact reduction I am looking for, it is an increase in smoothness. I know it can be done as I have seen youtube videos of the same footage I have that look better than what I can do. Like the one at the first post in this thread, which is still the best example I could find after watch probably hundreds of random interpolations.

I did read some of the thread you linked, but I didn't get chance to read all of it and I am not hugely knowledgeable about this so I didn't understand all of it.

[bradwiggo]

Maybe the GPU version of SVP works better than the CPU version. If you have a supported GPU try it.

I tried doing that using this script:

   # Header is exactly same as in previous example
   
   super=SVSuper("{gpu:1}")
   # Small 8x8 blocks with additional refine to 4x4
   vectors=SVAnalyse(super, "{ block:{w:8}, refine:[{thsad:1000}] }")
   # Conversion to 5/2 of source frame rate with 2nd SVP-shader.
   SVSmoothFps(super, vectors, "{ num:5, den:2, algo:2 }", mt=threads)

However, the results weren't much better.



I tried framerateconverter, but that was going to take 5 hours to do a 6 minute video file, whereas svpflow only takes 1 hour, why was it taking so long?

Framerateconverter script:

PluginPath = "C:\Users\bradw\Downloads\MeGUI-2836-32\tools\lsmash\"
LoadPlugin(PluginPath+"LSMASHSource.dll")
LoadPlugin("C:\Users\bradw\Downloads\masktools2-v2.2.17\x86\masktools2.dll")
LoadPlugin("C:\Users\bradw\Downloads\mvtools-v2.5.11.22\mvtools2.dll")
LoadPlugin("C:\Users\bradw\Downloads\RgTools-0.97\x86\RgTools.dll")
LoadPlugin("C:\Users\bradw\Downloads\MeGUI-2836-32\tools\avisynth_plugin\FrameRateConverter.dll")
LoadPlugin("C:\Users\bradw\Downloads\GRunT101\GRunT.dll")
Import("C:\Users\bradw\Downloads\FrameRateConverter-v1.0\FrameRateConverter.avsi")
file="source.m4v"
LWLibavVideoSource(file, cache=False)
AudioDub(LWLibavAudioSource(file, cache=False))
FrameRateConverter(NewNum=60, NewDen=1, Preset="slower", Prefilter=RemoveGrain(21))

[jagabo]

I opened the video I posted in my last comment in mpc hc, and the framerate was changing as it played, is this a problem?

Maybe. How much was it varying? MPCBE shows it varying from about 59.9 to 60.0. That small amount isn't a problem. But if you're seeing more you may have a playback problem.

[bradwiggo]

I opened the video I posted in my last comment in mpc hc, and the framerate was changing as it played, is this a problem?

Maybe. How much was it varying? MPCBE shows it varying from about 59.9 to 60.0. That small amount isn't a problem. But if you're seeing more you may have a playback problem.

It varies within about 1 fps. When I skip to a random point in the video it goes to just above 60 but then comes back down to 59.

I tried to interpolate with framerateconverter, but it was taking a really long time. It tales me about 1 hour to interpolate 6 minutes of footage with svpflow, but when I use framerateconverter it takes about 5 hours to do 6 minutes of footage, should this happen?

Also, would using the 1080p version of the video as the source help, as at the moment I am using the 720p version?