Improve the frame rate of your videos - amazing!

[Delerue]

I don't know if you have already heard about the revolutionary deinterlaced system called TrimensionDNM. It was implemented by Philips and used for the first time in WinDVD 6, a few years ago. Today the tecnology has a new name: TrimensionDNM Advanced. It appeared in WinDVD 7 and now in the last version of WinDVD, the 8 version.

How it works? It's simple. Philips explain it: 'Usually, the refresh rate of the PC is higher than the frame rate of the video source. Consequently, each input frame is shown multiple times by the monitor. This repetiition of the input frames introduces an unnatural, jerky motion, wich is called motion judder in the literature. TrimensionDNM solves this fundamental problem by calculating new frames between the originals'. So, that is: you come from 24 frames per second and go with real 60 frames per second. Oh, yeah, baby! Pretty amazing, I know.

Bellow we can see an image the shows the commom problem that we have with the repetition of the pictures that seems to double the image, tremble it and decrease its definition (at left). And the solution by Philips TrimensionDNM tecnology (at right):

http://www.trimension.philips.com/images/tiles.jpg

Believe me, the algorithm is amazing, and really works. The position of the objects of the entire scene is calculed in real-time, and you can't find any problem, even in high speed scenes (like the scene where Neo fights Morpheus, in 'The Matrix'). The most exciting thing is that all the movimentation becomes smoother and enjoyable to see. The pans (horizontal moviments of the camera) are the most beautiful thing, now with TrimensionDNM. Despite the fact that you can only use it with MPEG1 and 2 (i.e. DVDs), I can tell: if you like high FPS games you'll never want anything else; TrimensionDNM is the answer. :)

Here you can download a simple and small demo (579 kb) that can show the tecnology in action: http://www.trimensiontech.info/bin/TrimensionDNMdemo.exe

You can also download the original video to compare with your favorite player (WMP, VLC, MPC, BsPlayer, etc.): http://www.trimension.philips.com/index.php?page=demo_room.html

Or download the new WinDVD 8 Platinum trial version and test it in real world (i.e. with your favorite DVD). To enable the TrimensionDNM you must enter in the Video Center and then enable it manually, because it came disable by default. The only problem is that the new version that is available (8.0.6.101) maybe can't allow you to enable the TrimensionDNM. But if it occours, tell me and I'll upload another version to my server. Here is the direct link from Intervideo: http://www.intervideo.com/WinDVD/WinDVD_Download.jsp

Well, to conclude, I want to say that the tecnology isn't perfect in some specific situations. I think that WinDVD is really a great player, with unique features, but I have to tell that Intervideo is lauching unstable versions, and, worse, changing week after week the build and never tell to the users. Its website also doesn't have a forum. So, what really brings me here is that I found a strange problem with overlay videos running with TrimensionDNM. If someone has 'Contact' movie, can notice that TrimensionDNM blinks the image when the little girl, at a radio, says 'I need a bigger antenna', and then a morph to Jodie Foster looking at a bigger antenna. 'Lethal Weapon 2' has the same problem in a few scenes. I've already tried to send e-mails to Intervideo talking about this problem, but they are stupid and ask for my register number - that I don't have, of course. Hey! I'm trying to help them and they don't even listen to me because I don't have bought their product. I didn't buy it, and I'll never buy until they fix it. And qhat really drives me crazy is that the WinDVD 6 version doesn't have this problem. However, this version has others weird behaviours with TrimensionDNM, specially when you play dark videos, that tends to be shrinked constantly.

So, if someone could test the program deeply or know something about this little problem, I would love to talk about it.

Oh! I almost forgot to say that recently I discover a player that can do similar 'motion compensate' function. The name is Crystal (http://www.crystalplayer.com/index.php?page=downloads). It promisses to do the same thing that TrimensionDNM, BUT with any kind of video! The name of this tecnology is 'Motion Morphing'. I've already tested, but I noticed a few artefacts in some scenes. What you think, guys?

Thanks for your attention, and best regards.
;)

P.S.: sorry for my english; my official language is portuguese.
P.P.S.: right now seems that Philips website is down, so wait a few minutes and try again. ;)

[Shinigami-Sama]

I had this idea years ago
I thought it was called motion interpolation upscaling/supersampling
it seems simple enough to do, take the average between two frames untill you get the desired framerate

[Delerue]

I had this idea years ago
I thought it was called motion interpolation upscaling/supersampling
it seems simple enough to do, take the average between two frames untill you get the desired framerate

It's not that simple. You have to analise every object in the scene and calculate the future movement correctly. If it were simple, I think that Crystal and WinDVD would be a perfect programs, hehehe.

[quake74]

Well, the idea of interpolating two frames using motion estimation is what is used in the MVtools for avisynth, see MVFlowFps at http://avisynth.org.ru/mvtools/mvtools.html. Or at least, I kinda thinks that's how it's done...

[foxyshadis]

It's probably best to point to the actual heavy discussion on this topic instead of just "hey that sounds cool": http://forum.doom9.org/showthread.php?t=113256
The holy grail is still alchemist, which can do this in realtime in much higher quality than trimension, but it's a custom chip made to do nothing else.

I have heard good things about trimension, ranging from "doesn't suck too much" to how amazing it is. Unfortunately, lots of people have had problems with it, since windvd 6 at least; a quick search comes up with lots of "help me!" threads. Sorry I can't help.

[kumi]

If you can manage to present frames to the monitor in sync with the vertical refresh rate (with ReClock or Windows Vista et al), I see this as a rather unnecessary function. And wouldn't Trimension destroy the "film effect"? Granted some people might find that desirable...

[2Bdecided]

If you can manage to present frames to the monitor in sync with the vertical refresh rate (with ReClock or Windows Vista et al), I see this as a rather unnecessary function. And wouldn't Trimension destroy the "film effect"? Granted some people might find that desirable...

Too right!

It's same as (or similar to) the technology Philips put into their TVs. It unpicks 2-2 and 3-2 pull down and changes the frame rate to that of the display device. It does make film look like video. It also makes some really crappy low frame-rate sources look like video. I have even seen it help on bad frame-rate conversions to remove the choppiness!

IMO it gives much fewer artefacts than something like motion perfect, and is much faster than motion perfect or the AVIsynth tools.

If only there was a way of frame serving it's output, I would use it a lot! Heck, I'd even buy it!

Cheers,
David.

[2Bdecided]

btw, those "demos" are the same ones that were available several years ago. They're not demos of the actual technology at all - they just show you what it would look like, by taking a still photo graph and panning it at 24fps and comparing that to 60fps.

I haven't tried the WinDVD link yet. Is it easy to uninstall (and get back to what you had before!)?

Cheers,
David.

[KoD]

I wonder if Reclock works even with varaible frame rate material.

[Delerue]

Well, the idea of interpolating two frames using motion estimation is what is used in the MVtools for avisynth, see MVFlowFps at http://avisynth.org.ru/mvtools/mvtools.html. Or at least, I kinda thinks that's how it's done...

Would be great if I could understand something about this tool. :)

I tried to download AVISynth and MVTools, but it seems to much complicate to use it. Any ideas?

[Delerue]

It's probably best to point to the actual heavy discussion on this topic instead of just "hey that sounds cool": http://forum.doom9.org/showthread.php?t=113256
The holy grail is still alchemist, which can do this in realtime in much higher quality than trimension, but it's a custom chip made to do nothing else.

Same as above... :(

I have heard good things about trimension, ranging from "doesn't suck too much" to how amazing it is. Unfortunately, lots of people have had problems with it, since windvd 6 at least; a quick search comes up with lots of "help me!" threads. Sorry I can't help.

Yeah... As I said, Intervideo likes unstable versions. Really sad, because TrimensionDNM is almost perfect. The only little problem that I found is with overlay videos.

And what you have to say about 'Motion Morphing' by Crystal Player? It seems that it uses the same tool that you're talking about. Well, at least, I can see the same 'motion vectors' running.

[Delerue]

btw, those "demos" are the same ones that were available several years ago. They're not demos of the actual technology at all - they just show you what it would look like, by taking a still photo graph and panning it at 24fps and comparing that to 60fps.

The idea hasn't change too much. So, I think that these demos still have the point.

I haven't tried the WinDVD link yet. Is it easy to uninstall (and get back to what you had before!)?

Yeah, it's quite easy to uninstall WinDVD. Don't worry. ;)

[2Bdecided]

The idea hasn't change too much. So, I think that these demos still have the point.


Yes, but it's not a demo of the Trimension software itself. It's a program that moves a gif across the screen - nothing more, nothing less!


Yeah, it's quite easy to uninstall WinDVD. Don't worry. ;)

I'll have a go.

Cheers,
David.

[Delerue]

Yes, but it's not a demo of the Trimension software itself. It's a program that moves a gif across the screen - nothing more, nothing less!

No, no... If you download WinDVD you can see the same demo running inside, and you can enable or disable, increase or decrease the quality. You can also change the video file (to another MPEG2, I think) and see the same effect running in front of you. So, IMO these demos are real demos of TrimensionDNM tecnology. Of course, do the job with a picture seems to be less difficult than with a motion picture, but the idea and the effect is the same, and I think that the 'pams' are a good sample to show the amazing thing that this tecnology can do.

[Chainmax]

Speaking of the Alchemist thread, does anyone know if Helios (software equivalent developed by the same company) is out?

[JohnnyMalaria]

For DVD playback, it seems rather redundant.

Typical GOP-based MPEG2 streams contain all the motion vectors you need to be able to present the video at different frame rates.

The DNM is Digital Natural Motion - a software implementation of Philips' Natural Motion, introduced nearly a decade ago.

Of course, phase correlation remains the ultimate technique for motion estimation - as exemplified by Snell and Wilcox's Alchemist. (I wrote a phase correlation-based scene detector about 7 years ago.)

[Chainmax]

Would there be any chance to implement your scene detector in current Avisynth functions?

[Didée]

Ah, the PR departement did a good job. They promise you the moon, and another one took it for granted. (And it's an old hat by now, really.)

The obvious advantage of hardware based solutions is their processing speed, facilitated by the dedicated HW implementation (vs. the enormous overhead caused by an all-purpose system, a.k.a "PC"). Apart from that, software solutions like MotionProtectedFPS or customized variants thereof are not a bad alternative, especially when looking at the price. :)


Typical GOP-based MPEG2 streams contain all the motion vectors you need to be able to present the video at different frame rates.

Nada. Mpeg vectors have been achieved in order to minimize coding size. These vectors do not necessarily coincide with the "true" motion direction, and using them for motion interpolation is cheap, but not smart.

Of course, phase correlation remains the ultimate technique for motion estimation - as exemplified by Snell and Wilcox's Alchemist.
I might be leaning a bit far out of the window, but I think that phase correlation's "mighty-ness" usually gets overestimated.
Phase correlation in itself deals nothing. Its motion estimation is dislocalized, so it has to be combined with another kind of motion search (i.e. blockbased) anyways. Phase correlation only gives you the results that the localized motionsearch should find. But then, imagine e.g. that explosion scene with lots of flying debris: phase correlation will spit out "a thousand" targets ... and since you don't know which are which, and since the localized motionsearch will find lots of matches by itself anyways, the benefit is ... what? :)

To be clear, I don't claim that phase correlation would be useless. It's definetly a good "helper" in lots of the possible cases.
But sure as anything, it is not the one and only holy grale.

[JohnnyMalaria]

Would there be any chance to implement your scene detector in current Avisynth functions?

I'd have to dig out my old code.

I wrote it for something called "Rough Cut" - a scene splitter for DV format MOV (yes, Quicktime!) files that would then create a project file for the NLE I was using - EditDV. A great program - much respected but long since abandoned (originally owned by Radius, then Digital Origin and finally Media 100). It became known as Cinestream.

[Fizick]

I wrote a phase correlation-based scene detector about 2 years ago.
:)

Version 1.1, September 16, 2004 - Added experimental DepanScenes function.

[MfA]

Of course, phase correlation remains the ultimate technique for motion estimation - as exemplified by Snell and Wilcox's Alchemist. (I wrote a phase correlation-based scene detector about 7 years ago.)
It's pretty good, and it's truly amazing how they have been able to dominate the field (although I think they made a commercial blunder by ignoring consumer hardware). That said, phase correlation is not without it's faults. When there is significant amount of motion which isn't translation it has to fall back to primitive methods faster than absolutely necessary.