AviSynth or other way to remove blended frames from telecine film transfer.

[5stringsteve]

Hello all,
I'm new here, but this is a nagging old problem for many of us that work with film, or with film converted to video. I know about film scanning systems that capture frame-by-frame, but the system I'm working with operates in real time. The film runs at (round numbers) 20 fps with a 3-blade shutter, so there are 60 images per second. (The film advances every third time the shutter opens. This is how just about all film projectors work, though the number of shutter blades and frame rates obviously vary.) On the video camera side, I'm capturing 60p.

Just as a quick but relevant aside, my previous system was standard def, so it captured 60 FIELDS per second, interlaced into 30 FRAMES per second. The result was that typically every 3rd frame was a blend of 2 different film frames due to interlacing. The blending was mostly only noticeable with close-up hand-held camera movement. (Even Hollywood releases sometimes have frame blending, though it occurs every 5th frame due to the 24fps frame rate of the film with a 5-blade shutter)

So switching to the 60p camera system--which matches the 60/second film playback shutter--should have solved the blending issue right? Well not quite. It helped for sure. Each film frame is now represented as 3 video frames, as theory would predict, except that the third frame is sometimes blended. Because there isn't perfect synchronization between camera and telecine projector, a phasing occurs that gradually run in and out of perfect 1-to-3 film to video frame ratio. So for example video frames 1, 2, and 3 represent film frame 1; video frames 4, 5, and 6 represent film frame 2, and so on for several seconds until, say at video frames 91, 92, and 93--which should represent film frame 31--video frame 93 is also showing a ghost image of film frame 32--a blend of frames 31 and 32. (Or if the projector was running slower than the camera, video frame 91 would start to show a blend with film frame 30 still in the gate before the shutter closed, advanced, and displayed frame 31)
Using a higher shutter speed on the video camera would seem to alleviate this problem. However, the shutter speed must be a sufficient duration to span the entire period of the shutter cycle so as not to introduce another phasing problem--that of periodic brightening and darkening of the video.

So long story short I've done about everything I can to get the best results possible with the hardware of this system, and the results are impressive as is. But I wish to take this further if possible. I am also familiar with Avisynth and the very advanced filters and techniques that it promises. There is a plugin for AviSynth called SRestore that in theory as I understand should be able to compare all of my video frames and detect which ones are blended. Then it replaces that frame with whichever contributing frame is the stronger of the two. This would be ideal for the situation I describe because it would account for the dynamics of the phase mismatch.

The problem is that even with some help I have yet to get AviSynth working. There are missing dll files to install so that Virtualdubmod can install so you can run avisynth, then you have to install several plugins, then write or modify scripts. Then trial and error if it even works, and if not go to the doom9 forum. Really?? Isn't there anything else out there that can do this? Anything with a GUI so that you don't have to be a programmer just to do this fairly simple and common problem?? Or are there any AviSynth experts out there that could consult and set up a simple workflow for me??

Any insights, ideas, advice, or observations would be most welcome. Thank you all.

[Selur]

there are guis for Avisynth,.. in example AvsPmod (there are also a bunch of tutorials out there to get started with AvsPmod, in example the AVm tutorial on YouTube)
As a side note:
I haven't used VirtualDubMod for years and I do not see any good reason to do so, since it's old an buggy and hasn't been updated for years. So I would recommend to use AvsPmod or similiar and may be the normal Virtual Dub.

About your specific video, if you upload a decent sample and share it with the community there are probably a bunch of poeple who can help if you can't put the script together,...

[manono]

Yep, sample(s) please.

The problem is that even with some help I have yet to get AviSynth working.

Assuming you actually installed it write a simple one-line script:

Version

and stick it into a simple AVS file called maybe Version.avs. An .avs is simply a renamed TXT file. Then open it in VDubMod as you would any video (File->Open video file). If it opens and you see a little dark rectangle with the version of AviSynth you're using, then you got it working.

[5stringsteve]

Thank you both for responding. Yes, I will post a sample. That will make this all very clear and easy to visualize. I'll be back at my computer for that tomorrow (Monday), and should have something up later that day.

As for AviSynth, we did get that "working". It was Virtualdubmod that needed a .dll file for my computer and I gave up on it. The problem with AviSynth itself was first installing a bunch of required plugins and eventually the SRestore plugin that we were interested in. Dave, my tech friend managed all of that on his machine, but couldn't seem to get any results out of the SRestore script, except either error messages or some other funky video problem maybe with color shift, resizing or something. This was a few months ago so I don't remember exactly.

So I'm just revisiting this problem because there likely is a solution. I am not very adept with computers beyond what I manage to do with a GUI, but if I had a foolproof list to follow like a recipe, then perhaps there's an easy enough workflow for this for dummies too. When I first looked into this on Google a year and a half ago, I found and was blown away by the thread and Samples by VideoFred on the Doom9 Forum called The Power of AviSynth. So I've been very interested in implementing this technology ever since. It would be great to make it more accessible.

I am very interested to learn of AvsPmod. Thanks, Selur! I haven't heard of that one, so will find out what I can.
I'll get back with a post very soon. Thanks again.

[5stringsteve]

Hello again! I put two clips up in a Google Drive Folder to share. I hope it works!!!...

Clips:
https://docs.google.com/file/d/0B5Qv6PmJ4AF3OEVnQ0VaYkxVLTQ/edit?usp=sharing

Demonstration:
https://docs.google.com/file/d/0B5Qv6PmJ4AF3eFN2MmxUWTdNcVE/edit?usp=sharing

The first is a demonstration where I did a voiceover to some footage to try to explain exactly what is going on between the telecine projector and the video camera that causes the frame blending. It's not entirely necessary to see it, but I do slow the footage down to 5% speed so that the phase changes are easy to see. It's 5 mins long but does cover it thoroughly for those interested, and the young lady in it is very easy on the eyes. The other clip is 13 seconds, consisting of the 2 short clips used in the 5 minute demonstration. These should retain the original 60 fps frame rate and other characteristics.

Notes: The original files I'm recording are AVCHD 1280 x 720, 60p. This may be important when it comes to compatibility with whatever software/script might be used.

The demonstration video is an h.264 MPEG-4 file, so the frame rate is 29.97 For some reason the frame-by frame slow motion in the demo video looks like a series of dissolves, not instant jumps as did on my computer, but it still shows what it needs to.

A couple of further thoughts: My idea a few months ago was to use a plugin called Srestore, or maybe another one called RemoveBlend.
But a few minutes ago I no longer saw them listed at AviSynth.org. Maybe they no longer work with the most current version(?).
And this page appeared to be hacked: http://avisynth.org/mediawiki/Special:RecentChanges

Nevertheless, for the purposes of doing further processing in AviSynth such as grain reduction, dirt-removal, stabilizing and others that VideoFred(Freddy Van de Putte) and others have pioneered, I will most likely need a video file that has not only on blends, but no repeat frames at all--just a 1:1 film frame to video frame file. So to use Fred's, and maybe other scripts, I might need to first have a script or other process that will identify and extract all unique frames, while throwing out all repeat or blended frames. This might actually be easier than the process I describe using in the video demo and previous post where just the blends are replaced and the 60p file is otherwise left intact. At this point I'll take either one if it will actually work.

A sample of Fred's work is here: https://vimeo.com/11133342 I think you'll agree that the restoration possibilities make this worth pursuing. Thanks again for ideas or input.

Steve

[poisondeathray]

SRestore can be found here, the prerequisite filters and plugins can be found on the page
http://avisynth.org/mediawiki/Srestore

e.g

FFVideoSource("Frame Blend  Example Clips[0].mpg")
AssumeFPS(60000,1001)
SRestore(frate=20)

*Important that you use a frame accurate source filter, and encode linearly (start to finish) . Seeking can cause srestore to yield different , undesirable results each time you navigate

I recommend FFMPEGSource2 for the source filter
http://code.google.com/p/ffmpegsource/


Some of the dirt can be removed with various dirt removal filters, but the persistent long scratches (e.g. in the 1st clip) will be difficult to remove without some semi manual work

[Selur]

I recommend FFMPEGSource2 for the source filter

... or DGDecode if the input is mpeg-2.
btw. if I look for filters I normally check http://avisynth.org/mediawiki/External_plugins first

[5stringsteve]

Pure Magic!! Thank you, Poisondeathray. I downloaded your file expecting it to be a tutorial maybe, but sure enough it's a fixed version of what I uploaded! When I click frame by frame, the blended frames age gone!! Yeah! And the resulting file has an even 2 film frames per video frame throughout, regardless of whether the footage was in or out of phase. This is exactly what I was aiming for. The only difference is that I had imagined using "deblend-mode 3", where, according to the Srestore plugin page, "detected blends will be replaced with the neighbour that has the smaller difference" which would keep the 3:1 ratio of the 60p file intact. Of course that doesn't matter since I restored the proper playback speed easily in post @60% speed (66.6% gets it to the 3:1 original 60p speed, and 60% also corrects the 20fps transfer speed back to the original 18fps.) or @200% speed should get me the 1:1 ratio needed for the additional processing that Fred and others use.

So I am curious to know how that was done. Did you just use that bit of sample code you posted? I think not, but I'm pretty green when it comes to all of this. I've heard of ffmpeg, but don't even know what a source filter is. I will bring my tech friend into this so I don't start asking a bunch of dumb questions. He thinks that there needs to be some kind of host program that actually writes the output file. Maybe that's what AvsPmod does. Maybe something like Adobe Premier is needed. Anyway, we will get back to you in the next days with some hopefully intelligent questions. Thanks again to both of you!

[poisondeathray]

Pure Magic!! Thank you, Poisondeathray. I downloaded your file expecting it to be a tutorial maybe, but sure enough it's a fixed version of what I uploaded! When I click frame by frame, the blended frames age gone!! Yeah! And the resulting file has an even 2 film frames per video frame throughout, regardless of whether the footage was in or out of phase. This is exactly what I was aiming for. The only difference is that I had imagined using "deblend-mode 3", where, according to the Srestore plugin page, "detected blends will be replaced with the neighbour that has the smaller difference" which would keep the 3:1 ratio of the 60p file intact. Of course that doesn't matter since I restored the proper playback speed easily in post @60% speed (66.6% gets it to the 3:1 original 60p speed, and 60% also corrects the 20fps transfer speed back to the original 18fps.) or @200% speed should get me the 1:1 ratio needed for the additional processing that Fred and others use.

So I am curious to know how that was done. Did you just use that bit of sample code you posted? I think not, but I'm pretty green when it comes to all of this. I've heard of ffmpeg, but don't even know what a source filter is. I will bring my tech friend into this so I don't start asking a bunch of dumb questions. He thinks that there needs to be some kind of host program that actually writes the output file. Maybe that's what AvsPmod does. Maybe something like Adobe Premier is needed. Anyway, we will get back to you in the next days with some hopefully intelligent questions. Thanks again to both of you!

Srestore 'frate' parameter combines decimation, so it's getting rid of your duplicates as well, which - as you were already aware - required for farther processing . Duplicates will reduce the effectiveness of temporal and dirt/scratch removal filters

In a group of 3 duplicates (where one of them might be a blend), it picks 1 good frame. So you have 1/3 of the frames (all unique frames), and 1/3 of the fps, now 20 instead of 59.94 . This is effectively the same playback speed as the original. I suppose frate 19.98 would have been more accurate . 60000/1001/3 = ~ 19.98002

avspmod is a script editor, it can encode video as well. You can open avs scripts in something like vdub and use that to write and preview the output files

A source filter is something used to open the video in avisynth. There are different kinds of source filters, each with it's own pros/cons, idiosyncracies

[poisondeathray]

I think the actual film rate (projected) should be 18 FPS

EDIT: OK I misread your 1st post you just used "20" for round numbers

[5stringsteve]

Actually, by "round numbers" I meant 60fps instead of the actual 59.94. (20 is actually 19.98, and 30 is 29.97 technically to comply with broadcast standards.

20 fps (19.98) is the film playback speed that will sync with the 60p (59.94) video camera, yielding 3 duplicate video frames for every film frame. The file I got from you had 2 video frames of each film frame. I can work with any even number of video frames per film frame as long as there are no blends. So your process is fine. In your above post it says that "it picks 1 good frame". Great, I can work with that directly with Fred's scripts. But to say that that decimation restores the file to the same playback speed of the original is way off. It's confusing because there's the original film frame rate that was shot (18fps), the playback film rate (20fps), the initial video frame rate (60p--okay, 59.94), and then these derived video frame rates, which I prefer to think of as a percent of the project frame rate of the software timeline being used, which remains constant. So the picking 1 good frame decimation (if that's the correct word) of the 3 duplicate frames causes a 3x playback speed increase. When I work with my original 60p footage, I simply apply a playback speed of 90% to correct for the 20 fps transfer to correct back to the 18 fps original film frame rate. My timeline still represents 60 frames per second, but the software has added occasional duplicate frames to correct the speed.

Anyway, I'll be curious how many steps it will take to get from raw 60p transfer to the restored, finished file. If I understand correctly, most if not all processes can happen in one long series of scripts, but it's important to do them in the right order for best results.
Feel free to reply any time, but I will have some more pointed questions after I get my tech friend involved who can speak this language better than I can. Thanks.

[poisondeathray]

But to say that that decimation restores the file to the same playback speed of the original is way off.

note, the key word is "effectively" , and by the original, I should have clarified "original AVCHD recording", not the actual film recording

What I mean is 3x the framerate with 3x the frames composed of duplicates will look identical to 1/3 of the frame rate with 1/3 of the frames .

Anyway, I'll be curious how many steps it will take to get from raw 60p transfer to the restored, finished file. If I understand correctly, most if not all processes can happen in one long series of scripts, but it's important to do them in the right order for best results.

I recommend breaking out the srestore step to a lossless intermediate , because the temporal filters that come afterwrds can have problems when stacked together (frames mixed , out of order)

The file I got from you had 2 video frames of each film frame

You sure about that ? How are you determining this/ what software ?

Or do you mean 2 fields per frame ?

It should be 1:1 video frame to film frame

2 video frames per film frame would imply a duplicate frame remains per frame

When I work with my original 60p footage, I simply apply a playback speed of 90% to correct for the 20 fps transfer to correct back to the 18 fps original film frame rate. My timeline still represents 60 frames per second, but the software has added occasional duplicate frames to correct the speed.

You can correct the 19.98=>18 playback speed in avisynth as well

Depending on what types of things you are doing in the editor, and what your final format /delivery goals are - it might make sense to work in 18FPS timeline in your editor (you don't wan't to introduce blends, unless your delivery format requires it to make up a standard delivery FPS)

[manono]

My timeline still represents 60 frames per second, but the software has added occasional duplicate frames to correct the speed.

Which makes it what? Progressive 23.976fps? Progressive 29.97fps?

If the final output is to be a DVD, you can telecine off of19.98fps, which will require far fewer duplicate frames if the 'real' framerate is 18fps (or 17.982fps). And I'm not sure duplicating frames to get to 19.98fps (or 23.976fps) is the best idea though. I've seen retail DVDs inserting some few blended frames (Flicker Alley does this for silent films), and have experimented with it myself, both with good results. This is nothing your software can do, though.

[5stringsteve]

Hey guys,
I know what happened. I put the clip from poison deathray into my existing project, which is a 60p timeline. The clip is 30fps (again, using round numbers--really a 59.94 timeline, 29.97 file). So the software automatically displays a duplicate frame for each frame. A normal 30 frame timeline would show me the 1:1 clip. And yes, I am outputting to something standard--usually DVD, but it is always a standard frame rate of either 30 or 60. 60 is very useful for the reasons outlined in my video demonstration.

[poisondeathray]

The clip I uploaded was 20.0 FPS, because "frate" was set to 20

On a 20FPS timeline it would be 1:1 , on anything greater it should show either duplicates or blends

Is this premiere ? On a 29.97 timeline , it would introduce blends if you had frame blend on, duplicates if frame blend is off

[5stringsteve]

I use a program called MoviePlus by Serif. But I don't think that matters. The file properties told me it was 29.97. In my experience just about all files adhere to a fixed standard frame rate of 24, 30, or 60 and that's it. Same with the project settings in software. The frame rate of the file or timeline aren't actually changed, just a an adjustment by repeating or taking away frames to create the appearance of faster or slower motion. We're getting a little off topic from Avisynth, but if you have a different take on this than my years of experience, then I'd love to know otherwise. Thanks.

[poisondeathray]

You're right the framerate is irrelevant, its the actual frames (framecount) that is important

The point of using srestore is so you get the original film frames represented in video (ie. 1:1, no duplicates or blends) . It would be a shame if the program you used or settings you used re-introduced the very problems you're trying to get rid of

In most NLE programs, whenever the timeline FPS is > asset FPS, it will do what I described above - either introduce blends or duplicates to conform the asset to the timeline FPS.

Ideally you want to edit at the native frame rate, real 1:1 frames, because you can disrupt the cadence depending on the location of your cuts if it's not 1:1 - you might introduce blends or duplicates into the final product. The potential problems are analgous to editing progressive film DVD's with an interlaced timeline (combed transitions). All the other considerations should come later (eg. adding blends/duplicates or using a pulldown scheme for DVD)

[5stringsteve]

I adjust playback speed pretty often and am sure that it is just calculating and then evenly adding repeat frames or removing them as needed. No blends or other artifacts are introduced.

As I said, my current workflow is in 60p. I can simply adjust that file to 90% and be done with the blends still present. I've gotten better at minimizing the occurrence of them since the sample clips were transferred, so my newer files look great as is. If I decide to take the next steps and get into restoration beyond what my software already offers (good color correction and image adjustments, but mostly useless stabilization and other stuff) then it sounds like I'll do something like this in this order:
1. use Srestore to create a new 1:1 file.
Then with the new file:
2. maybe crop the 16:9 aspect ratio to match the film or 3:4 ratio depending on the film format (Super 8 & 16mm are a bit wider than 3:4)
3-10. work on shake/stabilizing, dirt, grain, color, sharpness, maybe something else, and finally speed correction.
And that process yields yet another file?
If so, I can specify the format I imagine?
And is using an AVCHD file to begin with okay for this? I could possibly get around that, but it would be neither convenient nor very affordable.

I'm so glad that you (Poisondeathray) were able to work with the file since one of the many variables I had suspected might have tripped us up was the color space of the file, which in AVCHD is YUV or 4:2:0 or something and I know that some filters will only work in some color spaces. So That's enough for now. I will try to get Dave (my tech friend) to at least try the decimation process in Srestore.
Does he need specific code or a script that's out there somewhere, or what? From what I can tell it must have to use a mode other than 1-5 which do not effect the frame rate. Again, mode 3 would be versatile for me since it would just remove the blends and not effect the frame rate/playback speed so that I could possibly use as is.
Anyway, thanks again for your tima and interest.

[poisondeathray]

1. use Srestore to create a new 1:1 file.
Then with the new file:
2. maybe crop the 16:9 aspect ratio to match the film or 3:4 ratio depending on the film format (Super 8 & 16mm are a bit wider than 3:4)
3-10. work on shake/stabilizing, dirt, grain, color, sharpness, maybe something else, and finally speed correction.
And that process yields yet another file?

You can combine 1,2,3 in a single script, many people do without problems. I tend to split out srestore in a separate step, I find when combined with other temporal filters the results can be dicey (mixed up frames and weird results).

If so, I can specify the format I imagine?

Yes. Avisynth is just a frameserver. It serves uncompressed video & audio . You feed that script to an encoder that accepts avs scripts. The encoder applies those changes in the script to generate a new video. The format can be bascially anything

And is using an AVCHD file to begin with okay for this? I could possibly get around that, but it would be neither convenient nor very affordable.

It's ok, but technically would be better if you used something less compressed

I know that some filters will only work in some color spaces.

Yes some filters only work in certain color spaces or have specific restrictions. It depends on which you plan on using

Does he need specific code or a script that's out there somewhere, or what?

The script is in the code box in my first post in this thread . You can tweak it however you wish - it's just meant as an example or starting point

From what I can tell it must have to use a mode other than 1-5 which do not effect the frame rate. Again, mode 3 would be versatile for me since it would just remove the blends and not effect the frame rate/playback speed so that I could possibly use as is.
.

What you're asking for is a the same 59.94 files, just minus the blends (replace the blends with an adjacent or duplicate). I don't agree with doing it that way, because you would still have the triplicates. All the cleanup, "videofred" stuff won't work properly (all the temporal filters, grain, dirt removal etc...) . But if you insist on doing it that way, during editing in your NLE you should pay attention to editing on 3 frame boundaries, if you do that, the cadence issues won't be a problem

[5stringsteve]

Cool. Wow, thanks. I totally get it that I would only use the 1:1 file for any filter processing. It's just that there may be occassions where I just want to remove the blended frames, do the 90% speed adjustment, and any conventional processing in post. Either way, it's very easy even in my software to change the 1:1 file to the 1:3 file or vice versa by adjusting playback speed to 300% or 33.333%.

I wondered if that example code you pasted in was the code you used, but it's so short, and doesn't appear to specify the kind of detailed instructions I would have expected. But that's just me.

I will work with Dave on this as soon as possible and let you know what happens. I believe he had it all functioning on his computer, but needed scripts since the modes or settings he used had no effect(?). It might also work better on my computer since I have a lot more processing power. I think we'll try this AvsPmod or maybe the regular VirtualDub. AviSynth.org recommended VirtualDubMod, but it was either buggy or I couldn't even install it. I'm not Sure if we can use Premier. Anyway, we'll do what we can for now.
Thanks again,
Steve

[poisondeathray]

Cool. Wow, thanks. I totally get it that I would only use the 1:1 file for any filter processing. It's just that there may be occassions where I just want to remove the blended frames, do the 90% speed adjustment, and any conventional processing in post. Either way, it's very easy even in my software to change the 1:1 file to the 1:3 file or vice versa by adjusting playback speed to 300% or 33.333%.

I still don't agree with doing it on a 59.94p timeline, because you will be outputing an interlaced 59.94i (or 29.97i, same thing, different naming conventions) DVD. You have less efficient encoding , more encoded duplicates, and lower quality . Moreover, the project will have hardcoded duplicates. This limits your options for other distribution formats. If you have a perfect 18FPS version, you can always add duplicates or do whatever tranformations later (e.g. what if you needed a PAL version, and you wouldn't want to upload a video with duplicates to web). IMO that 1:1 project should be a perfect master, not duplicate ridden . In project design and workflow you generally want to have the most flexibility and leave the most options open, keeping the best quality up until the end.

For NTSC DVD distribution the alternative way to what you're planning to do is to use soft pulldown method and encode a progressive DVD. And yet another method is the one videofred uses (motion interpolated new frames using mvtools). There are pros/cons to each of these methods, but they are discussed in more detail at Doom9

I wondered if that example code you pasted in was the code you used, but it's so short, and doesn't appear to specify the kind of detailed instructions I would have expected. But that's just me.

Yes, verbatim

FFVideoSource("Frame Blend Example Clips[0].mpg")
=> this is the source filter. It opens the video clip

AssumeFPS(60000,1001)
=> this fixes the framerate, because one of the "quircks" of FFMS2 is it sometimes gets the FPS fractionally off. It might interpret the fps as 59.91 or something. As you know, even "59.94" and "29.97" are approximations as well, the correct precise values are 60000/1001, and 30000/1001 . Unlike the settings you are using with editor , AssumeFPS alters the frame rate but doesn't change the frame count. (ie it's a perfect speed up or slowdown, no blends or duplicates are added, no frames are removed. It's analgous to "interpreting" the footage in a NLE, not conforming the footage ; conforming adds duplicates or blends or removes frames to adjust the framerate)

SRestore(frate=20)
=> This is calling srestore, "frate" is decimating to 20 FPS . Whenever parameters are omitted in avisynth, they are left at default values. Thus, only parameters that differ from default values need to be specified . On the srestore page, there are a bunch of other parameters you might adjust for blend detection, cache , etc.. The script provided is just meant as a starting point. In this case it actually picked 100% accurately. That's not always the case, you might need to adjust some of the parameters like threshold.


I'm not sure what other instructions you would like ? If you can't get avisynth started or preview scripts , just ask . Post the error message if any

http://avisynth.org/mediawiki/Main_Page#New_to_AviSynth_-_start_here

[5stringsteve]

Thanks again.
Thanks for the explanation of the code too. That should help demystify.

Good point about the 60p timeline. It does work well, and the DVDs look great. But I will look into both of those suggestions. Actually I think my software does create a progressive DVD. It definitely doesn't go blending frames to get to 29.97. All frames are progressive (or functionally progressive, since they derive from a progressive frame) and not blended except the original blends from the transfer.

I should have Dave joining the discussion shortly, though I think we have plenty to start working on...

[manono]

I still don't agree with doing it on a 59.94p timeline, because you will be outputing an interlaced 59.94i (or 29.97i, same thing, different naming conventions) DVD.

I don't either. Makes no sense at all to me. For most output formats the 'real' framerate is fine (17.982fps (or maybe 18fps) in this case). For DVD it has to be output as interlaced 29.97fps, but filling in the difference between 17.982 and 29.97 with dupe frames isn't the way to go. For something like YouTube where the lowest accepted is 23.976 (and 59.94fps isn't accepted either), other adjustments have to be made.

In my experience just about all files adhere to a fixed standard frame rate of 24, 30, or 60 and that's it.

And you have a film capture whose 'true' framerate is 18fps?

Okay, without doing any denoising, stabilization, removing the pincushioning, or any other clean-ups, here is a VOB file from a DVD I made of one of your samples. I removed the dupe frames, slowed it to 17.982fps, added a blended frame every 9 or 10 frames to make it 19.98fps again, encoded it as progressive 23.976fps and then slowed it again to 19.98fps by running the MPV through DGPulldown set for 19.98->29.97fps. In effect, this is a 17.982fps video in a compliant NTSC DVD. If you don't like the blended frames they can be replaced with duplicate frames, but in my opinion it plays more smoothly this way:

[poisondeathray]

For something like YouTube where the lowest accepted is 23.976 (and 59.94fps isn't accepted either), other adjustments have to be made.

Has something changed ? For YT, I have uploaded 1FPS before. Fractional values don't work, max is 30.0


RE: pulldown ; I know you've done a ton of these manono, I think that method you suggest is overall the "better" way . Of course pros/cons to each method

[manono]

Oh, when you re-download it from YouTube, it's still 1FPS? Or did they pad it out with duplicate frames? I maybe shouldn't have said 'accepted' but 'encoded', or something like that. Admittedly, I've never tested myself to see what exactly is going on.

[poisondeathray]

Not sure about the download, but it was from here
https://forum.videohelp.com/threads/348242-Can-someone-make-me-a-video-from-an-image

I just re-upped it to check. 1=>6

source
Bit rate : 2 415 bps
Maximum bit rate : 415 Kbps
Width : 704 pixels
Height : 400 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 1.000 fps

download
Bit rate : 42.1 Kbps
Maximum bit rate : 238 Kbps
Width : 634 pixels
Height : 360 pixels
Display aspect ratio : 16:9
Frame rate mode : Constant
Frame rate : 6.000 fps

framecount 541 vs. 3229 , can't tell if the added frames are blends vs duplicates on this type of static sample
filesize 165KB vs. 8.97MB , it added a dummy audio track, never seen that happen before

LOL bitrate INCREASED in the youtube version by 17x . When does that ever happen


//sorry for the off topic, carry on...

[5stringsteve]

Very interesting. Maybe not that last part, but that VOB file looked nice. Some blending, but not as objectionable like the original and looks a bit smoother. I'm just happy to know that the blended frames can be removed at all. As for pulldown and encoding, I will take a whole new look at that too. I hope to find something a bit less steps and conversion if possible. It seems that this might also be something handled in an Avisynth script or maybe not...

[manono]

Some blending, but not as objectionable like the original and looks a bit smoother.

Right. I first removed all the blending with SRestore. And as much as I detest blending myself, I purposely reintroduced it to bring the framerate back up to 19.98fps, the lowest framerate for NTSC that can be telecined. Rather than being a standard 3:2 pulldown, it's a smoother 3:3 pulldown. As I said, if the reintroduced blending bothers you, you can use a 'ChangeFPS(19.98)' instead to bring it back up to 19.98fps from the 'real' speed of 17.982fp. That will add in the duplicate frames required.

Very few of the silent films I've seen on DVD (or 8 or 16 mm films on DVD) with 'base' framerates below 23.976fps have used duplicate frames to bring them up to 29.97fps. Almost all hard telecine them. Nowadays we have a better way in that we can soft telecine them using DGPulldown. And if the real framerate is below 19.98fps, some sort of blending or frame duplication has to be used in order to bring it back up to 19.98fps.

It seems that this might also be something handled in an Avisynth script or maybe not...

You haven't mentioned anything yet that can't be handled in an AviSynth script. Any blend removal, clean-up, setting the right framerate, anything so far, can be done in AviSynth. You're familiar with VideoFred's work. He uses nothing but.

[5stringsteve]

Interesting... I don't understand the difference between hard and soft telecine, despite having read about it. Please don't bother explaining it here. I will find an article and share a link if I find it helpful. I'll brush up on these pulldown patterns too, since they seem to happen automatically in conventional software anyway, most of the time. Thanks again.

[manono]

If you've read up on telecine then you know it's how we go from (in the most common use of it) film to video - from the progressive 23.976fps of film to interlaced 29.97fps as required by NTSC DVDs and standard definition television broadcast. It's usually done by adding duplicate fields during playback. This telecine can be 'hard' with the duplicate frames/fields actually encoded into the video. Or it can be 'soft' with a set of 'flags' - think of it as being a kind of software - that tells how those duplicate fields are to be output as the video plays. In that case the video is encoded as progressive 23.976fps (most often) with the 3:2 pulldown (the soft telecine) added so that it outputs interlaced 29.97fps (or, more accurately, 59.94 fields per second).

Since the development of DGPulldown several years ago other framerates besides 23.976fps can also be soft telecined. It's perhaps most commonly used to take a PAL source, resize it for NTSC and, after encoding, apply a 25->29.97fps pulldown (or 3:2:3:2:2), to make an NTSC DVD. And it can be used for any 'base' framerate from 19.98fps up to 29.97fps, if the source is progressive. Thus it's also good to use for some framerates lower than film speed. For encoder efficiency, it's always better to encode at the lowest progressive framerate possible and apply a soft pulldown afterwards. And for smoothest playback, you don't really want to encode duplicate frames into the video if it can be helped. Playing back duplicate fields results in a smoother-playing video (on a television, say) than does playing back duplicate frames.