VFR for Fansub Encoders - how, why, WTF?

[Animeboy384]

120fps AVI
There's really no challenge whatsoever here. Use tritical's avi2tc package to get a decimated VFRaC raw containing all the frames, and a timecodes file (usage of the avi2tc package should be obvious). Encode said decimated raw, mux with timecodes. Simple. As far as I can tell, this should work with H.264 in AVI as well.

I tried to encode just like you told me, but for some reason, i don't get an audio out.

Sorry... do you have a reason as to why? =/

[Yumi`]

@Animeboy384
avi2tc only processes video, extract audio directly from the raw.



Forgot about an older tool: WMVTIMES.exe. It produces timecodefiles from wmv files. The latest version also fixes that weirdo first timestamp not being zero.
Also from its output you could confirm that wmv only stores timestamps with an accuracy of 1ms. Seems like mkv2vfr uses a 'diff' value on its own.

On another topic I rewrote my little toy.

Spoiler:

Vfrac(clip, string timecodesfile, int format) will decimate a hybrid cfr clip to its vfrac equivalent using a v2 timecodesfile. So it does the same as tritical's cfr2tc but:
* can be applied anywhere and anytime in your avs script.
* no useless vfrac raw is created on your HD.
* it is not limited to avi input
* !! it can only handle hybrids limited to one format. For mixed input (extremely rare) you'll need to trim the parts out and handle them separately.
Timecodesfile defaults to "2.txt", only v2 timecodes are accepted.
Format defaults to 1; 1 is NTSC (when you're dealing with a 119.88fps hybrid made out of parts of 119.88fps and partials), 2 is FILM (120fps and partials), 3 is PAL (50fps and partials).
Output is assumed at an arbitrary 24000/1001 fps.


Dupli(clip, string timecodesfile, int format) will create a hybrid cfr clip out of a vfrac by properly duplicating frames where necessary. More or less the same as tc2cfr but again no extra raw file and not limited to avi.


--------------------------

Example: decimate a hybrid 119.88fps raw.

Avisource("119.88fps.avi")
vfrac()

---------------------------

Example: create a 23.976fps cfr workraw out of a vfr raw

ffmpegsource("evilvfr.wmv")
dupli()
selectevery(5)
assumepfs(24000,1001)

---------------------------

Example: overlay cfr karaoke on a vfr clip
(Note that the karaoke needs to changed to 119.88fps before overlaying)

ffmpegsource("evilvfr.wmv")
dupli()
insertsign(last, karaoke.changefps(120000/1001), 3155, 6523)
vfrac()


--------------------------
--------------------------

Synchro(clip1, string timecodesbase, clip2, string timecodesoverlay, string timecodesout, float slack)

Still experimental and only advanced vfr users should dig in.
It outputs clip1 with duplicates frames on times where clip2 has a frame but clip1 does not.
Timecodesbase = v2 timecodesfile of clip1.
Timecodesoverlay = v2 timecodesfile of clip2.
Timecodesout = v2 timecodesfile of the output. (default = "synchro.txt")
Slack = defines the maximum time in ms between a new duplicated frame and an existing frame. It prevents creating duplicates too close to existing frames (equivalent to very high fps) and also prevents misses due to rounding errors. (default = 22.22ms)

Possible uses:
* apply something (karaoke or hardsubs fe) onto dedupped or low fps video without getting timing delays, missing lines and/or jerkiness.
* overlay something onto a clip with entirely different 'framerate' (cfr karaoke onto vfr clip fe) without missing overlay frames and/or inducing jerkiness.
* mux dedupped or low fps video with softsubs (current vsfilter only updates subs when a frame 'passes by' afaik).
* insert other crazy nonsense.

How to:
* clip1 is your base clip
* clip2 is karaoke/subs applied to an empty clip (with uncommon background color in order to catch any pixel change) which is then dedupped with threshold=0.0 .

-------------------------

Download here.

[DeSiBoY]

Is it normal for the video to be shorter in length after running it through avi_tc? I mean 24:59 to 21:30 is quite significant.....

[Yumi`]

If the assumed framerate of your resulting vfrac stream is higher than the overall average, then yes.
Most common example: if you have a hydrid 119.88fps avi consisting of 23.976fps and 29.97fps sections, then the (time) length of the decimated video will be longer if it is assumed at 23.976fps and shorter if it is assumed at 29.97fps. The higher you assume a constant framerate on a fixed amount of frames, the shorter the lenght will be (duh). Framerate has no meaning on vfrac streams and since framecount is constant, neither does time lenght.

[Wilbert]

Hello all!

I'm one of the Avisynth.org guys First, i would like to thank you guys for this guide!

Well, there's a rather large explanation of VFR at the Avisynth.org wiki, but it's mostly targeted at people who want to encode to CFR, recommends 120fps AVI as the "most compatible hybrid option" (which, IMHO, is pretty stupid), and doesn't mention such fansub-specific problems as AFX stuff or hardsubbing.

I disagree with your comment about the target people, but agree with the rest. The main reason that there is nothing about hardsubbing or AFX stuff, is that i don't have any vfr material with hard/soft subs to experiment with. So, i would be very grateful if you guys can provide me some example streams (some VOB and MKV samples).

Well, 120 fps AVI's are kinda like h.264 AVI's... it works on a normal Windows system, but it's ugly, breaks hardware players and is generally frowned upon.

I don't think that h.264 MKV will work either in those players Is it really true that hardware players can't cope with AVI's with drop frames? I always assumed they could, but perhaps i'm wrong?

I hope you can provide some samples, and in the mean time i will read the rest of this thread

[martino]

Originally Posted by Wilbert

So, i would be very grateful if you guys can provide me some example streams (some VOB and MKV samples).

Try Ayu's Gyagu Manga Biyori 2, that has quite a few 24/30fps scenes during the running of the whole episode, and not just as an OP/ED sequence (if the torrents/some other method of downloading isn't working and you're interested then I'll upload an episode or two -- they're pretty short/small anyway). I'm sure that there are many others, but no specific titles/releases can come up in my head at the moment. But I'm sure that others will know of some titles.

Not really sure what else, unless you don't mind if the only sections that have a different framerate than the main episode content itself are the OP and ED sequence... There are many of those, like True Tears, Mushi-Uta, Romeo x Juliet, Code-E, Ghost Hound for example (the ones that I remember anyway).

[Wilbert]

Try Ayu's Gyagu Manga Biyori 2, that has quite a few 24/30fps scenes during the running of the whole episode, and not just as an OP/ED sequence (if the torrents/some other method of downloading isn't working and you're interested then I'll upload an episode or two -- they're pretty short/small anyway).

I would be very grateful if you could upload two samples.

Btw, what's an OP/ED sequence? Opening and Ending?

Do you also have samples with other mixed base framerates (like 20fps and some other one)?

[D404]

Originally Posted by Wilbert

Btw, what's an OP/ED sequence? Opening and Ending?

Yes.

Originally Posted by Wilbert

Do you also have samples with other mixed base framerates (like 20fps and some other one)?

You really shouldn't see any 20fps sections at all unless you do some funky stuff. Really should usually only see 23.976 fps and 29.97 fps.

[PEM]

Originally Posted by martino

I'm sure that there are many others, but no specific titles/releases can come up in my head at the moment.

Sayonara Zetsubou Sensei is full of mixed framerates, all the way through.

[TheFluff]

Pretty much any .wmv is also VFR if you want a "sample" to test hardsubbing on VFRaC sources on. You could also just make a sample yourself from any video source with some simple Avisynth scripting and a handwritten v1 timecodes file.

Originally Posted by Wilbert

I don't think that h.264 MKV will work either in those players

You're missing the point; an important reason people want to use AVI is just because it works in hardware players. With 120fps it (frequently) doesn't, and hence using it for compatibility reasons is pointless. In any case 120fps is a horrible idea and I have no idea why you recommend it. MKV is easier to deal with (both on the encoding and the decoding side), is much more versatile, supports better codecs and arbitrary framerates, and unlike 120fps AVI it doesn't trick people into re-encoding the stuff to AVI's with 120 coded frames per second.

[Wilbert]

You're missing the point;

No, i don't.

an important reason people want to use AVI is just because it works in hardware players. With 120fps it (frequently) doesn't, and hence using it for compatibility reasons is pointless.

Ok, that's true.

n any case 120fps is a horrible idea and I have no idea why you recommend it. MKV is easier to deal with (both on the encoding and the decoding side), is much more versatile, supports better codecs and arbitrary framerates, and unlike 120fps AVI it doesn't trick people into re-encoding the stuff to AVI's with 120 coded frames per second.

Yes, in my original reply i said i agreed with your comment that '"most compatible hybrid option" (which, IMHO, is pretty stupid)', and will remove it. (I wrote that guide with several people, and didn't write that part.)
However i disagree with your assertion that 120fps is horrible and mkv/vfr is like heaven. I think that both have their advantages and disadvantages.

and unlike 120fps AVI it doesn't trick people into re-encoding the stuff to AVI's with 120 coded frames per second.

Sure, but re-encoding mkv/vfr is also very tricky for newbies.

When i write stuff on avisynth.org, i want to offer people some choices of doing things certain ways, and give pros and cons of them, unless a particular way is very bad for some reason. That's the reason why both methods are described.

[TheFluff]

Originally Posted by Wilbert

However i disagree with your assertion that 120fps is horrible and mkv/vfr is like heaven. I think that both have their advantages and disadvantages.

Oh, of course.
Disadvantages of 120fps AVI:
- doesn't support any other framerates than even fractions of 120 (or, well, technically you could have fun finding the smallest common denominator if you really really wanted to)
- comparatively large overhead
- cannot be easily remuxed
- guaranteed confusing, weird and/or completely broken results with a lot of software (as opposed to MKV where you usually at least get a honest "this isn't supported" answer); people just don't expect AVI to be VFR

Advantages of 120fps AVI:
- hmmm... it works in Windows Media Player without installing Haali's Media Splitter?

Obviously 120fps AVI has significant advantages and you would do well to consider using it in today's multimedia environment...


Actually, let me skip over all this condescending sarcasm since it's likely to get me banned again (sadly, the mods in this place don't enjoy flamewars like I do). Let me spell the facts out instead: while 120fps AVI might technically "work", there are no (zero, null, none at all) reasons to use it, EVER. Even the Windows Media Player "advantage" is kind of a stretch, considering the extreme popularity of VLC (which despite its obvious problems DOES support MKV and MP4) and the fact that as long as you're not using WMV/WMA in the AVI, it won't work in WMP without installing extra components anyway, and the average user in that case is highly likely to get some kind of codec pack, which in turn is highly likely to install Haali's splitter for him.

Originally Posted by Wilbert

Sure, but re-encoding mkv/vfr is also very tricky for newbies.

How is directshowsource("x.mkv", fps=23.976, convertfps=true) any harder to write than avisource("120fps.avi").selectevery(5)? Or if you want to preserve the VFR'ness, how is ffmpegsource() with the timecodes out parameter any trickier than fiddling with the tc_avi package?

[max2k]

Some encode newbie question, from some one who try to learn to it the right way.

Wich way would be the right to get an VfR timetable form a DVD iso? The fffpegsource() one, after riping the videostream seperatly (and later remurxin with audio) ? Any tips on the way?

On the 120fps xvid encode:
I think you would put the decoding PC to stress near an h.264 encode because of the framerate, why not use the better compresion of this codec and the vfr-bilty of mkv for an better quality release and xvid cfr 29.97 or 23.976 for an Standealoneplayer compatible lq encode. I dont know if a 120 fps encode is SPcompatible anyway, never tried it.

[Mentar]

Originally Posted by max2k

Some encode newbie question, from some one who try to learn to it the right way.

Wich way would be the right to get an VfR timetable form a DVD iso? The fffpegsource() one, after riping the videostream seperatly (and later remurxin with audio) ? Any tips on the way?

R1 or japanese R2 DVDs are all constant framerate 29.97fps, therefore VFR is not really an issue. It depends on what you do with it - if you run typical IVTC filters, the resulting framerate will be 23.976 fps afterwards.

In other words, you will need other means to turn a DVD source into VFR (e.g. via yatta).

[Quarkboy]

Originally Posted by max2k

Some encode newbie question, from some one who try to learn to it the right way.

Wich way would be the right to get an VfR timetable form a DVD iso? The fffpegsource() one, after riping the videostream seperatly (and later remurxin with audio) ? Any tips on the way?

On the 120fps xvid encode:
I think you would put the decoding PC to stress near an h.264 encode because of the framerate, why not use the better compresion of this codec and the vfr-bilty of mkv for an better quality release and xvid cfr 29.97 or 23.976 for an Standealoneplayer compatible lq encode. I dont know if a 120 fps encode is SPcompatible anyway, never tried it.

Use tritical's avisynth filter "TIVTC". There's a specific mode (2 pass) in the tfm function which will output a progressive stream and an mkv compatible timecode file.
Read the readme for details.
Note, it might require some manual tweaking, especially if there are some sections it detects as 60 fps instead of 30.

[martino]

Originally Posted by Quarkboy

Use tritical's avisynth filter "TIVTC". There's a specific mode (2 pass) in the tfm function which will output a progressive stream and an mkv compatible timecode file.
Read the readme for details.
Note, it might require some manual tweaking, especially if there are some sections it detects as 60 fps instead of 30.

That works as well... however, from what I remember that gives one hell of a different framerates mess when it generates the stream and the timecodes. I don't really ever remember it spitting out just 29.97 and 23.976 sections, no matter how much I tweaked it. But then, it's not like I've used it many times to make any general conclusion, so feel free to prove me wrong. There's also the modified Decomb... was called DecombVFR or something like that. However given that Decomb is a bit old, if you decide to choose this approach TIVTC would be the better choice (I think).

I'd go with what Mentar suggested, or if you don't feel like messing around with YATTA and don't mind a bit of manual work, you can always do it yourself. For example (note, I don't have access to AviSynth at the moment, so the syntax of some functions will most likely be incorrect and/or missing):

MPEG2Source("source.d2v")
TFM()
NoDec = last
TDecimate()
AssumeFPS(23.976)
Trim(NoDec,0,2000).AssumeFPS(23.976)+Trim(1600,0)

Where the first Trim takes 2001 frames from the non-decimated clip (defined before TDecimate), so that section is at 29.97fps and needs to be assumed at 23.976 in order for the splicing to work, since AviSynth does not support splicing of clips with different framerates. Second one takes frame 1600 and all the frames after from the decimated clip. Then you'll just have to generate the timecodes file:

#timecodes format v1
Assume 23.976000
0,2000,29.970000

This is just a simple example, and as you can see if you need more non-decimated sections it could get pretty messy, both scripting-wise and in terms of generating the timecodes file. Simply said, if you know how, YATTA can do this for you. If you don't, then you've been given enough choices by the replies in this thread.

[max2k]

Martino my first try on VFR encoding whould be a Anime you did for Ayu, Baccano! for a german fansubb. In the hoppe that i can release a nicer h264 encode with DVDraws this time.

I did it in with Xvid @29.976 fps because of the lowbitrate this piece of s...(only bitrate vice) work needs, even after filtering, including sharpening it like mad ,in a lossless export maraton (let me not lie but something like 14+ hours on my second rig for the lagarith), even learned to use share to get a littel better VfR mp4 raw. I encoded at this frame rate because i saw some ugly trembling at my first 23.976 fps encode in camaraswings/moving (whats the right spelling?) and action scenes. Baccano! was my first real released encode, i never heared some thing like "VFR for Fansub Encoders", but i thought this anime has to be real VfR (not only openning and ending) and the 29.97 encode was smoother overall in my eyes. The only one who would later have trouble wit hthis Decision whould be the typseter, me anyway, doing frame by frame typing on some moving shields. I got some talking behind my back frome some "more experienced" german fansub encoders, because my frist xvid encode in the scene was xvid 29.97fps 704*400 ( i heard some madman whispering over 16 bit is his relegion and why it is the rightway. (theFluff () around 130mb, beause "the onlyright way to do xvid fansub encodes whould be 23.976 fps, 704*396 around 170 mbs". I hope i did the right thing with my little, but growing knowlege...

At this time i wondered why Martino was only doing his h264 encode in MKV for this Anime, and get some leachers angry because they dont got their loved xvid version. Now i have the presumption, that youre encode could be VFR. I would reload one of ure TVraw encodes and one of the DVD ones to look in to it in the next days. anyway ( I have to reload because of Burglar, but thats another story)

Enough OT and is too moch mouch of my bad english grama anyway...

[martino]

Wait... what? Baccano was/is 23.976fps all the way through, both the DTV and DVD (R2) sources from my knowledge...

MKV was the best choice (if not only -- depends how you look at it) given the spec of the release.

[Mentar]

Originally Posted by Quarkboy

Use tritical's avisynth filter "TIVTC". There's a specific mode (2 pass) in the tfm function which will output a progressive stream and an mkv compatible timecode file.
Read the readme for details.
Note, it might require some manual tweaking, especially if there are some sections it detects as 60 fps instead of 30.

With all due respect, but unless this filter dramatically improved its performance in the last 6 months, it is (excuse me) useless crap. Let's face it, people, there is _no_ truly reliable way to properly determine VFR other than _using your eyes_ and making the proper call afterwards. I would strongly DISCOURAGE inexperienced encoders from using those semi-automatic "fire and forget" filters which pretend to mysteriously detect VFR parts, and afterwards even decimate to a VFRAC video with timecode file.

Typical mistakes made by TIVTC (but also by other forms of automatic VFR detection):

o Wild "flip-flopping around" framerates within normal 23.976 content (false positives)

o Misconstruing mostly-static scenes as VFR parts (false positives again)

o Tagging credits as unusable 60fps (what Quarkboy mentioned)

o Incorrect placement of "framerate change" markers, leading to strange jerk-slowdowns/speedups at the wrong spots


Now I don't want to put this filter down completely, it does what can be done automatically, about as good as I've seen it so far. But unfortunately, for hi-quality releases, it simply isn't REMOTELY enough. Also, since the output timecode file is basically always an erroneous mess, you are pretty much locked in to a pure MKV release - a proper avs scripting where you can create a correct CFR raw file by decimating the manually-adjusted 29.97 fps portions is not possible anymore, you need to use fps-changing tricks for that.

If you go down this path, be aware that you're only using very lightweight trickery, this is _not_ proper VFR encoding.

Instead, if you REALLY want to learn it properly, grab yourself a tutor and learn yatta.

[Wilbert]

Oh, of course.
Disadvantages of 120fps AVI:
- doesn't support any other framerates than even fractions of 120 (or, well, technically you could have fun finding the smallest common denominator if you really really wanted to)
- comparatively large overhead
- cannot be easily remuxed
- guaranteed confusing, weird and/or completely broken results with a lot of software (as opposed to MKV where you usually at least get a honest "this isn't supported" answer); people just don't expect AVI to be VFR

Your first disadvantage is not relevant. Most hybrid stuff is a combination of 23.976/29.97, and for that it works very well. Your fourth disadvantage is simply not true. Many software players can handle drop frames well. And yes, the other two are correct.

Advantages of 120fps AVI:
- hmmm... it works in Windows Media Player without installing Haali's Media Splitter?

It's advantage is that it does what it is supposed to do, and that they can be created using OS tools. And yes, before you continue with your flamewar again, MKV provides a good solution too. I never denied that, and i never claimed that i, personally, would use 120fps avi instead of mkv/vfr.

How is directshowsource("x.mkv", fps=23.976, convertfps=true) any harder to write than avisource("120fps.avi").selectevery(5)? Or if you want to preserve the VFR'ness, how is ffmpegsource() with the timecodes out parameter any trickier than fiddling with the tc_avi package?

I didn't say it is harder, but that's not the point. We often get questions, why the audio is out of sync when using

directshowsource("x.mkv", fps=23.976, convertfps=false)

or

ffmpegsource("x.mkv")

and encoding straight to XviD/DivX. It's not trivial to explain/understand what's going on in these cases.


But i didn't came here to start arguments like this. I just wanted to complement for the guide and get some samples with subs (which i received from someone).