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

[spiritgumm]

I have a similar goal/problem -- can I add to the thread since I think it's related?

I was trying to make an xvid from a silent film on dvd. Has alot of blends which I figured were permanent, and a duplicate frame every 5th frame, so I was only going to IVTC it. However, after doing that, the original looked slightly better. Then I read this thread, and tried "Srestore(18)" which vastly improved the video.

I wish the original dvd looked as good, so I was wondering how I could re-encode for dvd? HCencode wont output that frame rate (I was hoping to output 18 fps and run that thru DGPulldown).
clip.demuxed.m2v

[5stringsteve]

Hello, and yes I think this adds well to the conversation. I have been working with my tech and I think we have Srestore working as it should. The next step will be to decide on an output format for running any further processing/restoration scripts or just editing and conventional work in regular editing software. I don't think completely uncompressed AVI is an option. My file sizes need to be manageable, so Dave is looking into options for me that would be intraframe compressed. Maybe JPEG 2000? Anyway, he's planning on posting follow-up questions when he has specific ideas. So that's just an update (not much of one, but feel free to chime in anyone with suggestions on that)

But as far as your post, I'm very familiar with the every 5th frame repeat pulldown you get from a 24 fps playback to a 30fps video. And I think you're correct that the original frame rate was 18fps (or 16fps as were most early silent films). The overly fast motion also indicates that's the case. I use a system for those films that plays back at 15fps so that you get an even 2 video frames for every film frame, and find that it's better/easier to have too many frames to work with--i.e. play back too slow and "speed up" in post by removing frames, rather than playing back too fast and slowing down in post. But then again that system is perfectly syncronized and doesn't blend, whereas the blending is just due to the lack of synchronization of my other system.

Sorry that's not very helpful, but others here I'm sure can point us both in the right direction as far as outputting and encoding after Srestore.

Cheers,
Steve

[poisondeathray]

The next step will be to decide on an output format for running any further processing/restoration scripts or just editing and conventional work in regular editing software. I don't think completely uncompressed AVI is an option. My file sizes need to be manageable, so Dave is looking into options for me that would be intraframe compressed. Maybe JPEG 2000? Anyway, he's planning on posting follow-up questions when he has specific ideas. So that's just an update (not much of one, but feel free to chime in anyone with suggestions on that)

Commonly used lossless compression options for windows (large filesizes, about 1/2 the size of uncompressed YV12) : lagarith, huffyuv, ut video codec, ffv1 .

Performance wise, ut video codec (fastest scrubbing in editors). Compression wise, FFV1 (but slow as molasses)




Commonly used lossy compression options on windows: cineform, dnxhd, I-frame intermediates like matrox MPEG2 I-frame

cineform (has 5 user selectable quality levels, of course higher quality proportional to larger filesizes), there is a free version now, excellent timeline performance

dnxhd - but beware of gamma and levels shifts between different programs and platforms, MOV wrapped (many issues on PC) . Benefit is crossplatform compatibility (pc,mac,linux programs)

matrox - decodes as RGB, not YV12



My opinion - if you 're doing full scale restoration project, always use lossless

If slight quality loss is acceptible, I would go with cineform if your workflow is completely PC based . It's wavelet based compression , like JPEG2K

[spiritgumm]

heh Steve, there's one big difference in our situation - you're actually capturing the film. I'm working with the dvd someone already made. I've never heard of JPEG 2000. so I cant comment about that.
I tried 16 fps and 20 fps, but 18 fps appears to play best in my case.
Someone just reminded me of "Changefps(23.976)" which does what it says; keeps the timing the same by (I think) adding blends. The result isnt as sharp as I'd hope, but it's not as bad as the ghosting on the original dvd. Maybe one of the other guides will suggest a better solution.
Sounds like you've ironed out your conversion problems.

[poisondeathray]

Someone just reminded me of "Changefps(23.976)" which does what it says; keeps the timing the same by (I think) adding blends. The result isnt as sharp as I'd hope, but it's not as bad as the ghosting on the original dvd. Maybe one of the other guides will suggest a better solution.

ChangeFPS will add duplicate frames

ConvertFPS will add blended frames

The DGPulldown method will add duplicate fields using TFF/RFF flags to make it compliant with DVD 59.94i signal (also called 29.97i, same thing, different naming convention) - the difference being those duplicates are not encoded. Therefore, there is effectively more bitrate per frame (higher quality encoding)


There is no "best" solution here, it's all pros/cons. If you want smoother look maybe blends are better. 18 true FPS will be "choppy" eitherway

[spiritgumm]

poison, thanks for clarifying the difference between convert/changefps.
Did you download the clip [which I re-uploaded in the original dvd video]? Using Srestore(18) on it produces fairly normal video (as xvid). Are you saying dvd video would be choppy?
Know any free encoders to output 18 fps (which I would run thru DGpulldown after)?

[manono]

Then I read this thread, and tried "Srestore(18)" which vastly improved the video.

I'd do it for 17.982fps.

29.97fps is to 30fps as
17.982fps is to 18fps.

As you discovered, you can't encode for 17.982fps (or 18fps). 23.976fps is the lowest. In addition, if the plan is to encode progressively and then use DGPulldown on the lowest allowable framerate for NTSC (19.98fps), you'll have to get this thing from 17.982 to 19.98fps. You can use ChangeFPS(19.98) to introduce a duplicate frame every 9 or 10 frames, or do what I suggested and demonstrated earlier in the thread, and purposely introduce a blend every 9 or 10 frames. Up to you entirely.

You use the Motion.dll by mg262:

http://avisynth.org/warpenterprises/

like this:

SRestore(Frate=17.982)
BlendFPS(19.98,Aperture=0.15)#might have to adjust the aperature

Or maybe you'll want to interpolate the missing frames. You'll need both SVPFlow1.dll and SVPFlow2.dll

http://forum.doom9.org/showthread.php?t=164554

SRestore(Frate=17.982)
super=SVSuper("{gpu:0}")
vectors=SVAnalyse(super, "{}")
SVSmoothFps(super, vectors, "{rate:{num:20, den:18}}", url="www.svp-team.com", mt=1)#modded from a script I saw jagabo post in this forum

Using the interpolate frames method you'll find the results below (an MP4, not a DVD).

Now, how to encode for DVD? Add an 'AssumeFPS(23.976)' to the bottom of your script and encode as 23.976fps. Make sure to raise your bitrates by the same ratio (23.976/19.98=1.2). If, for example, you usually use a max bitrate of 9000, change it to 10800. When all done run it through DGPulldown set for 19.98->29.97fps. The length will remain the same as the original video.

[spiritgumm]

Then I read this thread, and tried "Srestore(18)" which vastly improved the video.

I'd do it for 17.982fps.

29.97fps is to 30fps as
17.982fps is to 18fps.

oh, right!
The dvd sounds abit tricky. I'll check out your suggestions tomorrow. Thanks!

[poisondeathray]

poison, thanks for clarifying the difference between convert/changefps.
Did you download the clip [which I re-uploaded in the original dvd video]? Using Srestore(18) on it produces fairly normal video (as xvid). Are you saying dvd video would be choppy?
Know any free encoders to output 18 fps (which I would run thru DGpulldown after)?

I meant "choppy" in the general sense , that 18 FPS will be choppy compared to , say 24FPS, or 29.97 FPS or 59.94p.

There is no good way to do this, because 18 isn't evenly divisible into 29.97 or 59.94 . So not only are the motion samples low, there will be a cadence judder as some frames will be on the screen longer than others (24p or 23.976 has it too, but most people in NTSCLand are "used" to it)

[poisondeathray]

@5stringsteve -

RE: intermediate codecs and 18FPS

If you're not sick of me trying to convince you NOT to use a 18FPS timeline - Another reason to keep it 1:1 frames is your lossless intermediate will be much smaller , easily 3x smaller in filesize than 59.94 with duplicates when using intraframe intermediate codecs. It's easy to convert to 59.94 using one of the various methods at the very end, but your master project and everything up until the end delivery should always be 1:1 frames

[5stringsteve]

Hi Poison et al.,
I thought you were urging me not to use a 60p timeline. I understand why that would be with all of those repeats.

The file you created has a 20.00 frame rate. I had never seen that spec. in a file before. I'd only seen frame rates that adhere to the 24, 30, and 60 standards (again, round numbers)

In my 60p timeline it is just like the original file without the blends (new, repeat, repeat; new, repeat, repeat; etc.)

In the 30p timeline it's new, new, repeat; new, new, repeat.

In a 24p timeline it's new, new, new, new, new, repeat...

All of these play back functionally at 20 fps and look great!

I don't have options for an 18fps timeline or anything other than those standard ones above.

I've been using 60p since that's the original frame rate of my video file. When I output to DVD the frame rate gets converted to 29.97 apparently by pulling out every other frame, so I'm not concerned about inefficient MPEG-2 encoding for DVD. In some cases, though, it does look like using a 24p timeline would be advantageous since it exhibits the fewest repeat frames to have to deal with.

Processing in Avisynth will be a different matter of course, where I will want to first create the 20fps (actually 19.98, right?) using Srestore to remove blends and all repeat frames as you did. As I said above, I can work with that file in my timeline as is, but more likely, since it now meets the 1:1 requirements of the AviSynth restoration scripts, I will run it through some series of scripts and then import back into a timeline and finish to DVD and whatever other format is desired.

So the intermediate output format from Srestore and other script processing has been a big question on my mind since the first time we tried this, my 13 second clip was over 600MB!!--about 1GB every 20 seconds, 3GB per minute!!

This is primarily amateur film footage were talking about, and yes, my original has already undergone AVCHD compression. I could and may choose to work around that by capturing from the HDMI out to a Blackmagic or other device using less compression. But I need to work efficiently and am more concerned with perceptual quality than some theoretical perfection that fills my hard drives, takes a lot longer, and ends up looking the same anyway.

So I'm looking for that happy medium that will probably be an intraframe codec. That will allow frame-independent processing of the 1:1 file, which should keep the scripts happy and able to achieve best results. I will look into your suggested codecs. Great to know that Cineform has a free version. I'm curious why JPEG2000 was not among ones you recommended. My knowledge of codecs beyond the basics is pretty limited though. I even posed this question to my tech:

True or False:
"Codecs are just like fonts; as long as you install them correctly on your machine, they will be accessible to any software that uses them."

I think this is true, but am less sure that if I output a file using say, a JPEG2000 codec in an AVI wrapper, that my consumer software, which works with AVI files such as DV-AVI, will be able to work with it.

Sorry for such a long post and different topics, but hopefully you all have simple answers to all of this.
Yeah right.

[poisondeathray]

Hi Poison et al.,
I thought you were urging me not to use a 60p timeline. I understand why that would be with all of those repeats.

Yes, and still am urging you not to.

The file you created has a 20.00 frame rate. I had never seen that spec. in a file before. I'd only seen frame rates that adhere to the 24, 30, and 60 standards (again, round numbers)

Remember you ran your projecter sped up at 20 , not 18 ?

That's why you use AssumeFPS(18) (or 17.982) at the end for the slowdown. Or you can do it in your editor if it is more flexible . The frame rate is irrelevant, it's the frame count 1:1 that is important. You can assume any framerate at the end

I don't have options for an 18fps timeline or anything other than those standard ones above.

Then that's a limitation of the software . Most editors will allow you to use (almost) anything

I've been using 60p since that's the original frame rate of my video file. When I output to DVD the frame rate gets converted to 29.97 apparently by pulling out every other frame, so I'm not concerned about inefficient MPEG-2 encoding for DVD. In some cases, though, it does look like using a 24p timeline would be advantageous since it exhibits the fewest repeat frames to have to deal with.

I would consider using different editing software. For all the reasons mentioned earlier . If that's not an option, then use 60p and cut on 3 frame borders. This way the cadence will be preserved (always triplicates)

Often when you do restoration, there is more than just running it through filters and/or NLE - there are other programs involved like photoshop, restoration programs, effects programs, compositing programs . When you transfer back and forth between programs you typically use a lossless intermediate. Yours will be >3x larger

Processing in Avisynth will be a different matter of course, where I will want to first create the 20fps (actually 19.98, right?)
using Srestore to remove blends and all repeat frames as you did. As I said above, I can work with that file in my timeline as is, but more likely, since it now meets the 1:1 requirements of the AviSynth restoration scripts, I will run it through some series of scripts and then import back into a timeline and finish to DVD and whatever other format is desired.

Yes, 19.98 . Then slowdown to 17.982

So the intermediate output format from Srestore and other script processing has been a big question on my mind since the first time we tried this, my 13 second clip was over 600MB!!--about 1GB every 20 seconds, 3GB per minute!!

This is typical for people that do restoration and video editing, you need more HDD's . It's up to you where you are willing to take shortcuts or quality hits

I'm curious why JPEG2000 was not among ones you recommended.

Because cineform is better than the JPEG2000 implementations I've used . Better quality and compression.

True or False:
"Codecs are just like fonts; as long as you install them correctly on your machine, they will be accessible to any software that uses them."

False; it depends on the software and specifics. For example a 32bit native codec won't work with a 64bit native program. Some codecs have both 32bit and 64bit installers . Some software can only access the VFW subsystem (video for windows), but there are other API's like quicktime (MOV)

I think this is true, but am less sure that if I output a file using say, a JPEG2000 codec in an AVI wrapper, that my consumer software, which works with AVI files such as DV-AVI, will be able to work with it.

Yes, it should be mostly true (VFW) , if you have matching 32bit vs. 64bit software and codec. (64bit vfw codecs are required for 64bit software)

[seronegative]

I suppose I should get in on this thread, being steve's "tech guy". I'm not much of one, & as Richard Dreyfuss' Close Encounters character exclaimed: "I have a million questions of my own"

For starters, is the capture methodology that steve's using ideal? I know about the numbers game, with 60 to 20; 3 to 1 & all that so it seems to make sense...the average codec, though, seems to pump out larger files than AVCHD & if srestore plumb works (back in december it did not, but TIVTC was not in the prereq list at the time) would not a smaller fps capture setting suffice, mathematically? My thought is to balance space & quality a little bit, esp. since avisynth srestore output is ideally uncompressed...

I guess I'll start with that before jumping into frameserving queries. Gratzi.

[poisondeathray]

For starters, is the capture methodology that steve's using ideal? I know about the numbers game, with 60 to 20; 3 to 1 & all that so it seems to make sense...the average codec, though, seems to pump out larger files than AVCHD & if srestore plumb works (back in december it did not, but TIVTC was not in the prereq list at the time) would not a smaller fps capture setting suffice, mathematically? My thought is to balance space & quality a little bit, esp. since avisynth srestore output is ideally uncompressed...

20 FPS would work fine for capture, but you're not going to find a consumer level camcorder or device that can record at that rate

Yes, avisynth output is uncompressed , many times larger than you're going to get with the orginal AVCHD recording (even with 'srestored' to 20p instead of 60p) .

You have other options for compression, but the trade off , of course is quality

Intraframe codecs are ideal for editing (fewer errors, fewer compatibility issues, "snappier" timeline performance), but poor compression compared to interframe (long GOP) compression such as AVCHD .

Another option would be to use h.264 (same type of compression as AVCHD) - you choose what settings you want ; you could make it as compressed (or not) as you want. You can select the GOP size, or intra if you want.

Personally , I wouldn't use anything lower than cineform at the lowest quality setting .

[seronegative]

The camcorder game is steve's gig... he uses the "prosumer" class camcorder (if there is such a thing). He'll pipe in on make & model when he gets around to it...there's a Panasonic & a Sony & I do recall 24P as an option last I got a demo (not sure about 20). For my purposes, the main issue is workflow for steve.

Taking your advice as I understand it, srestore is implemented and outputted before graduating to "the fred stuff". Vdub to uncompressed works like a charm but I cannot find a way to apply compression via vdub (not your job to tell me on that one--but feel free). After reading the frameserver faq, I dragged & dropped an avs script into Miro Video Encoder installed on my system on a whim & voila, thumbnail & output options. Very few of them. I work with Premiere (which steve is not fond of) & I know about the plugin... My question is, would we be better off frameserving to steve's app of choice, or splitting workflow using multiple encodings (with a terabyte drive, I believe he has space for uncompressed output)...

in the FAQ lingo: should we use a proxy utility or go direct & can you recommend applications. Thanks so much for your help. I've learned more from this thread than hours of documentation research.

[poisondeathray]

The camcorder game is steve's gig... he uses the "prosumer" class camcorder (if there is such a thing). He'll pipe in on make & model when he gets around to it...there's a Panasonic & a Sony & I do recall 24P as an option last I got a demo (not sure about 20). For my purposes, the main issue is workflow for steve.

20FPS isn't standard. You're not going to get 20FPS available as a recording option until high end pro cameras, variable frame rate recording options like varicams. > $20K

Taking your advice as I understand it, srestore is implemented and outputted before graduating to "the fred stuff".

This step is a must, non negotiable. Dupliates just destroy the effectiveness of other temporal filters and "videofred stuff"

Vdub to uncompressed works like a charm but I cannot find a way to apply compression via vdub (not your job to tell me on that one--but feel free).

1) load your avs script
2) video=>fast recompress (use this to prevent colorspace conversions, staying in YV12)
3) video=>compression , select the compression and settings (you need to install a codec e.g. cineform would appear here if you had it installed)
4) file=>save as avi

There are other encoding options, like using vdub's external encoder feature, but stick with that for now

After reading the frameserver faq, I dragged & dropped an avs script into Miro Video Encoder installed on my system on a whim & voila, thumbnail & output options. Very few of them. I work with Premiere (which steve is not fond of) & I know about the plugin... My question is, would we be better off frameserving to steve's app of choice, or splitting workflow using multiple encodings (with a terabyte drive, I believe he has space for uncompressed output)...

I would not frameserve with srestore to other applications . I would use a lossless intermediate and split off at least 1 step. That is my personal opinion; some people combine multiple steps. The danger is srestore will lose it's place (frame accuracy) when you do non linear seeks (scrubbing) . My advice is to split off at least that srestore step definitely before editing, and probably before the video fred stuff

1TB won't be enough for this project, unless the duration is very short

in the FAQ lingo: should we use a proxy utility or go direct & can you recommend applications.

I would not use a proxy unless you have older hardware

Proxies are usually only if your hardware isn't up to snuff (timeline performance is too sluggish) . Otherwise you actually end up using more disk space (original assets + proxies) . This is where intermediate choice makes a difference. Something like FFV1 will be unbearable to work with, but small filesizes (for a lossless codec).