Star Trek Hybrid Challenge!

[Guest]

The Star Trek series is notoriously difficult to encode due to its hybrid content (mixed progressive 30fps/progressive 24fps/video). With a view to stimulating creative thinking about how to deal with hybrid material I have posted this challenge, which will remain sticky until the conclusion.

Here is the source VOB contributed by MrBunny (right click and Save As):

http://shelob.mordor.net/dgraft/misc/TNGsample.vob

My humble attempt at encoding is here (right click and Save As):

http://shelob.mordor.net/dgraft/misc/trekdivx3.avi
or
http://shelob.mordor.net/dgraft/misc/trekdivx5.avi
or
http://shelob.mordor.net/dgraft/misc/trek.mpg

The challenge of course is to make the best encode at 23.976. The chief criterion is the smoothness of the film and video segments. Please post your entries here.

At the end we will all reveal our methods. Have fun!

[Si]

What flavour of divx have you used Donald - your avi just crashes my player (MS MP6.4) :(

regards
Simon

[Guest]

DivX 5.02.

Have you tried VirtualDub, BSPlayer, or anything else?

Bad download?

Do you want a different format? If so, please specify it.

[Si]

I don't have DivX5.

MS MPEG-4 V2 would work for me but I wonder now what a good standard format for doing compressed samples would be - I stopped using DivX when it moved to 4 (could never get the early versions to work :( )

Maybe re mpeging it to MPEG-2 (or even -1) would ensure maximum compatibility?

regards
Simon

[Guest]

I just upped a DivX 3 before reading this. :)

Give me 20 minutes to up an MPEG1.

EDIT: OK. See the link above for the MPEG1 file.

[Suzahara]

A very nice, unchoppy sample. Have you tried this on other parts of a Star Trek ep? Or even a full ep?

[Si]

Ta Donald
Si

[MrBunny]

Hey,
Not trying to be contrary or anything (I appreciate all of Donald's work), maybe more of a devil's advocate. If you check the first few seconds, the outer edge of the Enterprise-D's saucer seems to fluctuate a bit during the pan. I don't really know how to describe it, but it's probably something to do with the interpolation. I think it's even more noticeable (than it'd normally be) due to the contrast of very light grey with the black of space.

Otherwise it looks great, nice and smooth. Any secrets you want to share with us perhaps? It might help people think of ways to improve on it.

Mr. B

P.S. I know it's not possible for it to be "perfect", but I'm just giving feedback :)

[resonator]

It's a nice thing to play with these sources, btw. Babylon 5 is the same beast, but I guess it doesn't make any sense to invest this amount of time to get it done on >160 eps of Trek. Or did you figure out a way to do this automatically?

Why not simply use decomb to get 29.97 progressive frames per second and live with it?

[Suzahara]

Originally posted by MrBunny
Hey,
Not trying to be contrary or anything (I appreciate all of Donald's work), maybe more of a devil's advocate. If you check the first few seconds, the outer edge of the Enterprise-D's saucer seems to fluctuate a bit during the pan. I don't really know how to describe it, but it's probably something to do with the interpolation. I think it's even more noticeable (than it'd normally be) due to the contrast of very light grey with the black of space.

Otherwise it looks great, nice and smooth. Any secrets you want to share with us perhaps? It might help people think of ways to improve on it.

Mr. B

P.S. I know it's not possible for it to be "perfect", but I'm just giving feedback :)

Which version did you look at? Didn't seem to notice anything like that in the divx5 :confused:

[Guest]

Originally posted by Suzahara
A very nice, unchoppy sample. Have you tried this on other parts of a Star Trek ep? Or even a full ep? No, Sir, as I have only the VOB provided to me. But I have no reason to think it would be any different.

[Guest]

Originally posted by resonator
but I guess it doesn't make any sense to invest this amount of time to get it done on >160 eps of Trek. Or did you figure out a way to do this automatically? Why not simply use decomb to get 29.97 progressive frames per second and live with it? How do you know how much time was invested? Your point is well taken, Sir. But I am trying to make a point here. It will become clear soon enough. ;)

[Guest]

Originally posted by MrBunny
Not trying to be contrary or anything (I appreciate all of Donald's work), maybe more of a devil's advocate. If you check the first few seconds, the outer edge of the Enterprise-D's saucer seems to fluctuate a bit during the pan. I don't really know how to describe it, but it's probably something to do with the interpolation. I think it's even more noticeable (than it'd normally be) due to the contrast of very light grey with the black of space.
Perhaps slightly, yes. But compare it to mode=3, which seriously exhibits the effect you mention. It's like night and day.

[MrBunny]

Originally posted by neuron2
Perhaps slightly, yes. But compare it to mode=3, which seriously exhibits the effect you mention. It's like night and day.

Yes, I agree completely. My own attempts at that scene were definately worse than your result. I just wanted to point that out in case it hadn't been noticed.

@Suzahara

I watched the divx3 one initially, but the same effect can be seen in all of them. Just watch the tip of the Enterprise-D's saucer section. I know it's very picky, but again, just trying to point out the aspects that can be improved.

@neuron2

If you want a longer sample to experiment with, just holler :)

Mr. B

[aharden]

I'm going to try this out since I have a better handle on Telecide and Decimate than I did when I originally tried to encode my Season 3 TNG DVDs. Should be fun...

[lancer]

well that seemed pretty smooth to me Neuron. There were no obvious jerks in the movement, of course, apart from picard moving across the bridge there are no big movements that would glaringly stand out.

the wibble everyone mentions on the saucer of the enterprise seems to be more to do with the original VOB than the encode to me.

[MrBunny]

Originally posted by lancer
the wibble everyone mentions on the saucer of the enterprise seems to be more to do with the original VOB than the encode to me.

If you were watching the VOB on the computer with a software DVD player, then most likely that the wibble would appear. However, I would assume that on TV it wouldn't be so, though the poor guy that I am, I can't afford a DVD writer to test that theory. :(
The wibble is consistant with the type of manipulation I think Donald is playing with.

I guess using xvid's interlaced setting with TV-out might work to test for the wibble.

@neuron2
So what is your secret? ;)

Mr. B

[Guest]

@MrBunny

I rendered the clip once using MotionPerfect to convert to 23.976 and once using Decomb. Then I used Trim() to combine the appropriately rendered video and film portions.

The point I am making has actually already been made by resonator. The amount of time and effort you want to invest will have the biggest effect on the outcome.

I do agree that mode=3 looks poor with smooth video pans. But mode=1 looks poor with smooth film pans. :(

If these things bother you and you have the time and energy, buy MotionPerfect and do it as described above. For a single movie, OK. For 160 episodes, well, you'll need a lot of energy. :)

[McQuaid]

I was disappointed in that your means of dealing with this problem included MotionPerfect. I was very pleased with your results when I watched the sample, and was anxious to see the solution. I thought, finally Star Trek can be done properly. Now even though it was not implicitly stated, I thought this was an avisynth challenge (or a decomb challenge to be more precise), and not by any means necessary.

Your solution is very creative but as you said, it would take a great amount of time to tackle a project such as this. Heck, even just capping a few favourites and using this method would be very time consuming.

So, I'm a little disappointed that, short of this method, there isn't really an idea way of dealing with this using decomb.

Now, I realize this isn't the fault of your great work on decomb, I guess ideally, we would need a method of controling frame rate within a video file, but as far as I know that's not supported by any compression method.

Now, just to get a better understanding of hybrid. Are there two types of hybrid video? My original understanding of hybrid clips is simply separate video and telecined film sources in one clip. But I believe I read on another post that with star trek, sometimes effects that are interlaced are laid on top of an already telecined source, which makes it very difficult if not impossible to ivtc correctly. Is this accurate?

I wonder this because I have captured other shows that are hybrids such as Oz, and granted the results are not perfect but I was very pleased with the results and using the hybrid method in decomb leaving the frame rate in tact. Yes there would be that 5th ghosting image on the the telecined parts, but overall everything was smooth. But with Star Trek, I seem to have nothing but problems. Pans seem jerky, sometimes even people just talking or a turn of a head looks choppy. I even tried encoding one scene of voyager with 2 people talking in a shuttle. I went through it and could tell the scene was just telecined video. I enocoding using the default deomb method bringing the frame rate to 23.976 for testing purposes just to see if that scene could look right, but even it looked choppy. Now, I'm capturing from tv, so they are not dvd sources, but I capture using avi_io (picvideo 19), just about never have frame drops, and use my vcr as the tuner to get the best reception.

So, what is your current decomb setting recommendation for hybrid clips (keeping 29.97 frame rate) capped from slightly noisy analog source?


Also, is there a list of hybrid shows? It annoys me when I think something is pure telecined only to realize that its actually hybrid. Family Guy, for example. I thought it was just a telecined cartoon, but for some quick pans its not. Btw I kept one of these clips with the pans, I have no where to host it right now but here is an unaltered pic. Just to test it, I just deinterlaced the scene but it still looked wrong, so I don't think using decomb's Hybrid method would help here. I was low on disk space, so this particular cap of family guy was picvideo on 17 I think, so maybe that accounts for the difficulty with it.

But back to your solution, maybe what we need is a motion perfect like plugin for avisynth coupled with decomb?

[aharden]

I thought it was a AVISynth/Decomb challenge, too. Oh well. I did get clean 24fps for the initial pans of the Enterprises, but the solution for that caused Decimate to want to blend frames in the bridge sequence. It wasn't as clean as what Donald produced.

I was going to post a particularly challenging clip from the episode "The Ensigns of Command". There's a part where the Sheliak ship is seen leaving on the main bridge viewer and it's a 30fps interlaced effect, I believe. I remember that scene being particularly ornery when I was trying my hand at encoding it at 24fps.

[Billy|B-A]

****...messed up. Posted a bad encode. sorry.:scared:

[cube]

Dunno about caps but family guy season 2 PAL DVDs were ****** to deinterlace, Telecide(guide=2) (no decimate) gave the best results, albeit still with a ***** of ghosting.

The unaired ep on the dvd (I thought it was tame/respectfulish enough) wasn't interlaced, why couldn't they have all the eps like that :(

MODERATOR: Edited to remove profanity and other objectionable content. Follow forum rules or strikes will follow.

[jfcarbel]

neuron2
How did you get the VOB DVD video into MotionPerfect since it does not support AVS files?

Can you explain in more detail how you used MotionPerfect. I am confused why you would use Decomb after using MotionPerfect since all the frames would now be at 23.976.

[resonator]

Originally posted by jfcarbel
Can you explain in more detail how you used MotionPerfect. I am confused why you would use Decomb after using MotionPerfect since all the frames would now be at 23.976.

Well, I can tell you what I did for testing purposes (did a PAL transfer) on an Anime that contained some hard to process scenes.

For 29.97fps progressive:
-Used decomb in order to get 30 progressive frames per second
-Used VDub in fast recompress mode to make a temporary clip; used the HuffYUV codec.
-Opened this clip in Motion Perfect in simple mode and set the target framerate to 23.976 (or 80%), left the target duration at 100%.
-Set Motion Perfect to High Quality.
-Edited the processed scene back into the rest of the episode using Trim().

For 29.97fps interlaced:
-Used SeparateFields(), SelectEvery(2,1) and BicubicResize(720,480) to discard the bottom field, resize to full frame size and therefore get "progressive" frames.
-Then I did just that same as described above.

Finally I ended up with an ultimatly smooth and totally motion-jerkyness-free transfer at 23.976fps which I speeded up to 25fps and resized to 720x576.

[Guest]

Originally posted by jfcarbel
neuron2
How did you get the VOB DVD video into MotionPerfect since it does not support AVS files?

Can you explain in more detail how you used MotionPerfect. I am confused why you would use Decomb after using MotionPerfect since all the frames would now be at 23.976. Using Avisynth and VDubMod, I transcoded the VOB fragment to HUFYUV. I then made two clips from it: 1) Used MotionPerfect to reduce the frame rate to 24fps. 2) Used Decomb to IVTC. Finally, I used Avisynth and Trim() to patch together appropriate sections from the two rendered clips.

[jfcarbel]

Just to clarify - you used MotionPerfect on the interlaced parts to get them to 24fps and then used IVTC on the film parts and then combined them. Is this correct?

Also couldn't one use MotionPerfect on the whole movie? I know it may be more time consuming but it would make it easier from the point where you would not have to find the interlaced parts and cut, process, and paste back. So for a job that you run overnight I assume MotionPerfect should do just as good of a job as IVTC on the film parts or am I wrong in thinking this?

[Guest]

Originally posted by jfcarbel
Just to clarify - you used MotionPerfect on the interlaced parts to get them to 24fps and then used IVTC on the film parts and then combined them. Is this correct? Yes, after deinterlacing the interlaced parts.

Also couldn't one use MotionPerfect on the whole movie? I know it may be more time consuming but it would make it easier from the point where you would not have to find the interlaced parts and cut, process, and paste back. So for a job that you run overnight I assume MotionPerfect should do just as good of a job as IVTC on the film parts or am I wrong in thinking this? MotionPerfect frame rate conversion is not an IVTC process.

You're welcome to submit an entry for the challenge if you think you can do it better. :)

[mf]

What about this (http://home.bip.net/gunnart/video/deshaker.htm) ?

[Guest]

Originally posted by mf
What about this (http://home.bip.net/gunnart/video/deshaker.htm) ? I give up. What about it? It appears to have no relevance to this thread.

[mf]

Originally posted by neuron2
I give up. What about it? It appears to have no relevance to this thread.
From http://home.bip.net/gunnart/video/deshaker.htm
makes panning, rotation and zooming smoother

Isn't that what you want when you convert the non-film parts? I mean, IVTC makes nonfilm choppy. So this ought to fix that.

[resonator]

Originally posted by mf
Isn't that what you want when you convert the non-film parts? I mean, IVTC makes nonfilm choppy. So this ought to fix that.

The approach is different, while MotionPerfect tries to capture the movement within the frame and with the information tries to "render" the exact same movement at at different framerate without having to cross-blend pictures, the filter you mentioned does what a lot of camcorders are able to do too, stabilizing the movement of the camera itself. For instance: You shot your video handheld and got a shaky video out of it. Imagine giving the camera to a person that drank too much coffee - this filter tries to fix that.

Deshaker - fixing motion on the x and y-Axis (height, width)
MotionPerfect - fixing motion on the z-Axis (time).

[mf]

Originally posted by resonator
Deshaker - fixing motion on the x and y-Axis (height, width)
MotionPerfect - fixing motion on the z-Axis (time).
I assumed it did the z-axis too. Sorry then.

[resonator]

I think there's no need for an excuse. ;)

[Si]

I was just wondering if anyone has (and would mind posting) the PAL R2 equivalent of the test .vob.

It would be interesting to see what the big girls and boys did to it to get it to 25 fps.

regards

Simon

[Fasola_Jan]

Hi :)

Nothing important, really...

They just destroyed it.

(Shots with brightness enhanced from
VD with ViewFields filter enabled,
ST TNG, PAL, 1x17, beginning sequence))



(how can I post more than 1 picture?)

Fasola

[Guest]

Now you know where all those blended fields come from. :)

[Si]

Its obviously a good technique then ;)

[fccHandler]

My contribution to the challenge:

http://home.attbi.com/~blade66/TNGtest.avi

It's certainly not a general "solution," but I had an idea to kick up the frame rate to 120 fps, repeat each NTSC frame 4 times, and repeat each FILM frame 5 times. Also, if the NTSC fields were bobbed (instead of deinterlaced, as in my example) it should be possible to reverse the process and recreate the original hybrid sequence with minimal loss.

But I hope you have a fast computer (faster than my P3 650, anyway) to play this thing back.

[MrBunny]

Actually I'd been thinking about 120fps encoding recently myself. Considering the nature of DCT encoding, perfect duplicate frames should take "zero" space except the header. Therefore in actual fact, the 120 fps is just the size of the distinct frames + 120 frame headers. In my one really quick experiment, it seemed that the size (quant2 xvid) of a 120fps encode (29.97 ivtc'd to 23.976 and then interleaved with itself 5 times) would be close to that of a decimate(mode=1) encode. I think this is an idea worth exploring. The search for "hybrid" perfection will never end.

As a sidenote, this would kinda fall into the domain of the suggestion I made a while ago about having conditional filtering based on decomb's film vs ntsc metric. If film, decimate, then copy 5 times, if ntsc, copy 4 times.

Any opinions? neuron2? :D

[bilu]

The main problem is playback on older machines, in my laptop PIII-500 it starts skipping even at 48 fps! :scared:


Bilu

[JohnMK]

That's not the main problem for most of us.

[bilu]

I know ;)

[bilu]

Problem solved :)

1) Speed up the IVTCed parts from 24fps to 30fps - Avisynth needs constant framerate during the whole stream.
2) Do a timecode file using the format below
3) Load timecode file and the encoded AVI into MKVMerge and voilá! a "Variable Framerate Perfect Hybrid Encode" :)

Yes, it's that easy ;)
MOVIE.AVS

Loadplugin("D:\vfr\tools\Decomb510.dll")
Loadplugin("D:\vfr\tools\MPEG2Dec3dg.dll")

function FOrder(clip c) {
fo= GetParity(c)? 1:0
return fo
}

function DClip (clip c){
oldfr=c.framerate
fo=FOrder(c)
c=Telecide(c,fo,guide=1,post=2)
c=Decimate(c,cycle=5,quality=3,ovr="")
c=assumefps(c,oldfr)
return c
}

mpeg2source("D:\vfr\sources\TNGsample.d2v",idct=2,cpu=0)
trim(0,297)+DClip(trim(298,0)TIMECODES.TXT

# timecode format v1
Assume 23.976
0,297,29.97MKVMERGE

mkvmerge -o "D:\vfr\encodes\movie.mkv" -d 0 -A --timecodes "0:D:\vfr\encodes\timecodes.txt" -S "D:\vfr\encodes\movie.avi"Using this MKVMerge version: http://www.bunkus.org/videotools/mkvtoolnix/win32/mkvtoolnix-0.7.5.rar
UPDATED

And as Gizmo pointed out on this thread:
http://forum.doom9.org/showthread.php?s=&threadid=64314Please note that this requires the latest 1.0.1.8 mkv splitter from the Guliverkli project http://sourceforge.net/projects/guliverkli/ for playback.
Please also note if you use mpc to disable the built in MKV splitter.For those who don't know the MKV (Matroska) format, you don't need to play it under Gabest's MPC. Zoom Player or WMP9 work fine after installing the MKV splitter too.

EDIT: It's online at http://moodub.free.fr/matroska/vfr_startrek.mkv

Best regards,
Bilu

[Wilbert]

I would like to see the vob-file. Could someone upload it somewhere?

[bilu]

It's a 35080KB VOB file. I'm trying to find someone to upload it.

Bilu

[bilu]

@Wilbert

http://isabelle.math.ist.utl.pt/~l52768/TNGsample.rar

Original VOB. (EDITED URL)


Bilu

[lighty]

Now if only somebody could make some avisynth filter to automate that process. My guess is that telecide procedures in Decomb filter could be used in some way (I am just guessing though) to detect different framerates and to create some kind of timeline markers for MKV muxer?

neuron2 took sabattical from his video work so I hope to see anybody take a look at this method. :cool:

[vhelp]

I remember D/L'ing this way back in January I think. I messed around
w/ it for a while, but lost track of happenings on this. At one point,
I thought I had it solved, but I'm not too sure now, that I had a 2nd
look at the sample I did. But, I think the answer pretty simple. In
my opinion, culprit is roge Field swap during certain scene changes.

For some strange reason, the "Field" get swaped during two scenes in
the sample TNG vob, accoring to my analysis..

A - the one where scene switches from Picard looking out at the viewscreen
B - just before it pans to the viewscreen and he's moving inside it.
.
.
And, the 2nd time it Field swaps back to normal, when it:

A - switches back to focus at Data.

My guess is a lot of Field swapping:

* during major scene-changes or,
* During the special BLUE-screen panning in/out.

Either way you look at it, these are those "dirty" cuts that that gets
incorporated into the "digital" encoding and authoring process for DVDs.
To rule this out, if anyone has a VHS version of that same episode, they
should not have faults w/ any IVTC patterns. Anyways..

If the above is correct, then all that is needed would be a "marker"
for just those Field swaps locations.
.
.
I wonder if they are garbage, and can be deleted (w/out effecting the
audio sync)

Oh, back to my analigy..
Yeah, after those two Field swaps (above) even my IVTC went back to
work w/out an disterbance in the pattern. hmm.. now that I think of it,
I think the answer is clear. At those Field swaps, just re-swap them!! :eek:

A - http://www.westnet.com/~vhelp/im/picard.swap.viewscreen.jpg

B - http://www.westnet.com/~vhelp/im/viewscreen.swap.data.jpg

Any response to my ideas ?? :D
-vhelp

[bilu]

@vhelp

I haven't found any problems with automated filters such as Telecide/Decimate, Uncomb/Decimate or SmartDecimate. The only problem was the hybrid nature of it for which there is a solution now.

But you probably know this and this answer won't help you :rolleyes:


Bilu

[Si]

@vhelp
Some (maybe the majority) of old VHS recordings MAY have simpler IVTC requirements.

So a simple and fast IVTC process MAY work :)

This is probably due to the fact that old films/filmed series had their special effects (if any) added in at the film stage and the whole programme/film was simply converted to video.

As technology progressed, things like ST TNG had all sort of processes done to it inc film-->video->special effects added-in and spliced together with pure video as well.

Luckily (at least as far as us ST fans go) the effects seem to be going back to the film (24fps) stage and life is getting easier :)

(Unless your in one of the european countries where they seem to have extra special NTSC-PAL convertors that can mess up the simplest of transfers :( )

regards
Simon