FILM9 - Restoration software for old films (8mm, Super8, 16mm) and video

[boffee]

Could be part of the latest Windows 10 Version 1909. Don't know for sure, just guessing.

[Gelinox]

I'm also on this version of Windows 10.

I just did the following test:
- Uninstalling the 1909 update
- Uninstalling Microsoft C ++ .....
- Restart
- Download and install version 1909

The Microsoft C ++ 2015-2019 package was not installed.
Strange ... but that doesn't change my previous advice.

[fabiosan]

First of all thanks for Film9 it is an amazing tool and is doing wonders to my Families 40year old memories.
But I seem to have found an issue when using a 29.97 FPS source of a 18FPS Super8 original.
When setting up the project I inform it was scanned without sync and the original is 18FPS.
The issue reproduces even with all filters turned off:

The first pass seems to generate a complete though accelerated output in chutier/ClipPasse1.avi, and the second pass somehow gets the wrong size for the input video and crops the end of the movie which is also accelerated by a ratio suspiciously close to 18/29.97.

Everything works fine if I set the input frame rate to 29.97 but then I get duplicted frames and the dust removal doesn't work as well on duplicated frames.

I tried different sources and different machines with the same result.

Merci,
Fabio.

[GillesH]

Hello Fabio,
First, confirm the capture method you used.

For the WITHOUT SYNCHRO option, the speed of the projector must be adjusted to 1/3 of the speed of the video sensor.
- Standard PAL / projector speed = 16.6 fps
- Standard NTSC / projector speed = 20 fps

Please check and confirm this important point. This WITHOUT SYNCHRO option allows you to eliminate 1 image out of 3 to find image-video = image-film.
And so, the end result is dependent on this capture method.

For the WITH SYNCHRO option, no duplicate images are needed for the Clip-Source.
Your clip must have 1 video image = 1 film image.

[fabiosan]

Thanks Gilles, I naively gave my films to Costco(grocery store in the US) for scanning thinking they couldn't mess this up...
YesVideo does the conversion for costco and their website doesn't clarify their methods.

What I can assert is that the video has 29.97 FPS and only 18 of those are unique, which is easy to tell due to sheer amount of dust and scratches on this specific clip.
so of the first second I see 30 video frames that map to these film frames:
0,1,1,2,2,3,4,4,5,5,6,7,7,8,8,9,9,10,11,11,12,12,1 3,13,14,15,15,16,16,17...

From your answer it seems Film9 is making a ratio between film and video to chose which video frames to drop.
The confusion is whether each of these fps parameters come from...
Is Argentique Film Turning Speed the fps of the film or the video(scanner)?
I assumed it was the film but your message seems to imply it is the scanner speed.

If I put 29.97 there the output gets funky with frames out of order like, 0,1,1,2,2,0,3
If I put 30, partial success, the film frames get all singled out and all make out to the output, which is now 3x speed up.
This is all without any filters on, just pass through.
Turning on interpolation doesn't help fix the speed.

[fabiosan]

So turns out they dint' all made out, only 11 of the first 18 frames came out when selecting w/o sync 30fps.
The frames who made out are:
1,2,5,7,8,9,11,12,13,15,16,

So 0,3,6,10,14,17 got dropped

[GillesH]

Thank you Fabio for this information.
It is very clear and therefore I confirm that the fact that there is duplication of images brings an abnormal result.
Very often, companies that digitize old films only copy the original, transforming it into video standard (NTSC or PAL).
And so, an original film of 18 fps ends up with 29.97 fps and therefore with duplicated images.

And so, you should definitely not use the WITHOUT SYNCHRO option with FILM9.
You can use the WITH SYNCHRO option by indicating that the original speed is 29.97 fps or use the VIDEO TAPE option.
This VIDEO TAPE option will automatically take the speed of the clip.
But, since there are duplicated images in this clip, not all the corrective filters will be fully operational.
In particular, the task CLEANING filter. This system works in temporal mode and compares each image to each other.
If 2 successive images are identical, the dirt will not be deleted.
That is why it is absolutely essential that all the images are different and correspond to the original film.

For these films digitized by industrial scanners, there are several AVISYNTH solutions to restore the original speed and therefore delete duplicate images.
Among these solutions, there is a plugin called "TIVTC" which has several filters (TFM, TDecimate and others). This removes duplicate images.
On this VideoHelp or on Doom9 forums, you will find more detailed explanations by specialists.
But, there are also other filters to do this job.

To conclude, with the options WITH SYNCHRO or VIDEO TAPE, you can use FILM9 and most of the correction filters will work.
Like the Colors, Degrain, Sharpness, etc. settings, there is only Cleaning task which will not be fully operational.

All my encouragement for the future.
Cordially
Gilles

[ultradj83]

Dears,
running version 3.00.

Gettijng in touch with this software I attempted to add a film digitized to JPG with a frame by frame scanner.
Well, JPGs are 1804x885, 200dpi.

I'm returned this error:


[Attachment 52358 - Click to enlarge]

Not sure what does this mean.
Maybe the problem is that height (885) is odd?

Regards

[sholee]

You can't divide 885 by 2. Cut one pixel from the height and it will work

[ultradj83]

You can't divide 885 by 2. Cut one pixel from the height and it will work

Well, yes, I'm searching for a way to do this losslessly...

[JoeSchmo999]

I get this after just installing Film9 and running one small film clip. I installed Fourcc after getting the error once. Didn't help still get error.
avisynth open failure Avisource: coundn't locate a decompressor for fourcc lags
C:\program files(x86)\film9\IFO\Dll\sl.avs, line5

[manono]

Lags is short for the Lagarith codec. Install it and try again.

[JoeSchmo999]

Yep that did it. Now to see how good I can make the film look. Thanks

[robertzombie]

Hello,

I've been having a look around the Film9 software, hoping to use it to restore some commercial VHS tapes. However, I'm having an issue whereby the video it displays in VirtualDub is covered in blocks/squares. These squares are seen on the processed video file too.

The clip plays fine in VirtualDub without Film9 (see screenshot), so I'm wondering if I've not configured Film9 correctly?

The clip is just something I quickly trimmed using VirtualDub, I realise the audio is not useable.

Thank you,
Robert.

[GillesH]

Hello Robert,
Indeed, I see the problem that we have never seen before.
A priori, this comes from an encoding problem which is not well accepted.
I offer you a solution that I just tested.
From your Videos, you should do a preliminary encoding of your Videos in VirtualDub.
Take "Video / Compression / Lagarith Lossless codec".
Then, you go back to FILM9 with this new encoded clip and there, there is no more problem.
That said, I advise you to deactivate "AutoGain" in "Colors Base", the correction is too strong.
But you may find other settings.
Good luck.

Edit : for Audio, don't forget to choose a format in the Projects window (MP3 or others).
And, increase the Volume in "Pre-Settings / Audio"

[robertzombie]

Hello Robert,
Indeed, I see the problem that we have never seen before.
A priori, this comes from an encoding problem which is not well accepted.
I offer you a solution that I just tested.
From your Videos, you should do a preliminary encoding of your Videos in VirtualDub.
Take "Video / Compression / Lagarith Lossless codec".
Then, you go back to FILM9 with this new encoded clip and there, there is no more problem.
That said, I advise you to deactivate "AutoGain" in "Colors Base", the correction is too strong.
But you may find other settings.
Good luck.

Edit : for Audio, don't forget to choose a format in the Projects window (MP3 or others).
And, increase the Volume in "Pre-Settings / Audio"

Thanks, Gilles. That clip came from a transfer workflow I was using at work. I'm just in the process of setting up a capture chain at home and I'm planning to record video to HuffYUV AVI and audio to 16/48 PCM WAV in VirtualDub. Would the resulting file be compatible with Film9?

[GillesH]

Normally, FILM9 supports the Huffyuv Codec (fourCC = HFYU) and Audio PCM. It should work.
Thank you for confirming us.

[robertzombie]

Film9's filters are quite remarkable!

[fabiosan]

Thank you very much Gilles and the whole Film9 team.
It did wonders to my families memories originally stored in mute super8.
Here is one of the videos with side by side comparison.

The only challenge I had was to extract the 18 unique frames our of every 29.97 that the scanned video had.
I tried many methods but FFMpeg was the one I had most luck with, in case anyone needs it here is what I run:
FFmpeg -i input.mp4 -vf mpdecimate=max=1:hi=1400:lo=700:frac=0.6 -vsync vfr out.mp4
From there I imported the move as with synchro 18fps on Film9 and it did a great job.

I also used FFMpeg for interpolating the video to 60fps as it did a better job than the Film9 solution.
FFmpeg -i input.mp4 -vf minterpolate=fps=60:mi_mode=mci:mc_mode=aobmc:vsbm c=1:me_mode=bidir out.mp4

If I may make a few suggestions:
1- have a way to run Decimate or UnTeleCine to the input video as many of these cheap grocery store scanners leave us with sub par frame rates.
2- Apply Cleaning for dust and scratches BEFORE the color balance as it seem to affect dust detection.

Merci beaucoup!
Fabio.

[johnmeyer]

Thank you very much Gilles and the whole Film9 team.
It did wonders to my families memories originally stored in mute super8.
Here is one of the videos with side by side comparison.

The only challenge I had was to extract the 18 unique frames our of every 29.97 that the scanned video had.

If the film transfer house delivered the results as 29.97 telecined video, they should be shot. There is no need to do that any more.

In the future, I would suggest that you use the TIVTC filter for AVISynth for removing pulldown. It is, by far, the best tool for this that I've ever found.

The actual script depends on how many pulldown fields/frames your transfer house added.

Once you get the unique frames, you may want to re-time the speed at which they play. There is no standard for 8mm and 16mm amateur silent film, and I've found that really old hand-cranked 16mm can be as low as 10-12 fps. Most 8mm and 16mm is 16 fps. Occasionally some is as fast as 18 fps. Only sound film, or silent film shot on sound equipment, is 24 fps.

Here are the key lines in the script I use to recover 16 and 18 fps material that has been telecined to 29.97 fps.

tfm(display=false)  #consider adding cthresh=255
tdecimate(display=false,mode=0,cycleR=2,cycle=5) #18 fps  # 3:3:4 field count
#tdecimate(display=false,mode=0,cycleR=7,cycle=15)#16 fps

[fabiosan]

If the film transfer house delivered the results as 29.97 telecined video, they should be shot. There is no need to do that any more.

Yeah, the internet tried to warn me too, but you can't beat the convenience of submitting films for scan at to the grocery store that you already go to, I figured I could recover from whatever shit they did.

Thanks for the suggestion, I actually tried it, but given the speed of the projector and the scanner were not in sync it ended up eating up some frames every several seconds or so, I figured that actually visually looking for distinct frames irrespective of speed was a better idea since camera, projector and scanner could be out of sync...
And it worked, at worse I got a few duplicated frames due to extra dust or them being half way out of the shot, enough that made it look like different frames.

[johnmeyer]

First of all, I hope the transfer equipment they used was "frame accurate," meaning that, immediately after the transfer was finished, but before they added the pulldown, there should have been exactly one frame of film on each frame of video, with no blends of adjacent frames, and no flickering from a projector shutter. If that was not the case, there is other recovery software you can use which can get you close to a frame accurate transfer. I can provide a link to that if you suspect you have blended frames.

As for TFM/TDecimate not working, that usually happens if you don't set the Cycle and CycleR parameters correctly. To do that, you should break down the video into fields (because pulldown is usually done at the field level for 29.97 output). Then, go through a few seconds of video that contain fast motion, and do it field-by-field. Count the total number of fields while also counting the total number of unique images. These two numbers can be used to give you the Cycle and CycleR parameters. For most telecine added to film transfers, the pattern is usually quite reliable and doesn't change or skip because it is done with a simple algorithm without reference to scene changes. However, if you suspect any sort of interruption, you should double or triple the Cycle and CycleR values so that the software analyzes a larger group of frames when making its decimation decisions. This is often all that is needed to get perfect results.

For virtually all the work I've done (I've done this on probably a hundred films, or more), this software works perfectly, with no errors that I've ever seen.

[manono]

Better might be for fabiosan to post 10 seconds of the DVD or MP4 or whatever it is, from a section with steady movement.

[johnmeyer]

Better might be for fabiosan to post 10 seconds of the DVD or MP4 or whatever it is, from a section with steady movement.

I agree, especially because the side-by-side video he posted looks absolutely awful (really bad encoding artifacts) and there is no way to know how much of that happened when preparing for upload, and how much is "baked in" to the video he received from the transfer house.

[GillesH]

Thank you very much Gilles and the whole Film9 team.
If I may make a few suggestions:
1- have a way to run Decimate or UnTeleCine to the input video as many of these cheap grocery store scanners leave us with sub par frame rates.
2- Apply Cleaning for dust and scratches BEFORE the color balance as it seem to affect dust detection.

Hello Fabio,
Thank you for your explanations and, in particular, for these FFmpeg methods.
It is true that there could be, still, modifications to apply to FILM9 and your suggestions are very interesting.
But our small team is only made up of 2 persons, well occupied elsewhere.
For several years, Roland and I have tried to make the use of AVISYNTH filters easier with this FILM9 interface.
It took a lot of work. In all modesty, we are only passionate amateurs.
The project is dormant, but we remain tuned for possible advice.
Greetings.
Gilles

[fabiosan]

For several years, Roland and I have tried to make the use of AVISYNTH filters easier with this FILM9 interface.

And you did it Gilles, Film9 is immensely helpful.
I'm also a dev and make my own coding contributions to a better planet on my spare time, so I get it.
I made suggestions as a mere attempt to contribute, without any expectations.

First of all, I hope the transfer equipment they used was "frame accurate," meaning that, immediately after the transfer was finished, but before they added the pulldown, there should have been exactly one frame of film on each frame of video, with no blends of adjacent frames, and no flickering from a projector shutter.

Oh absolutely not the case, the pulled down frames are never identical just very similar, and yes some are blended, some show half a frame as it moves out. From what I can tell they just pointed a HandyCam at a wall where they projected the movie with a crank projector
At least that is the word on the street.

Better might be for fabiosan to post 10 seconds of the DVD or MP4 or whatever it is, from a section with steady movement.

Thanks guys for wanting to chase this, I really got what I needed with mpdecimate but if you really want to, here is short clip:
https://photos.app.goo.gl/44R1A9z9LFgLwoDr9
I guess it can serve as a tale to why not scan your films at Costco/YesVideo.

Cheers,
Fabio.

[robertzombie]

I'm having an issue with the Colours Bars > Auto White function. Occassionally it'll produce short bursts of static like this:


[Attachment 52411 - Click to enlarge]

Any thoughts on why this might be happening?

[johnmeyer]

Using FILM9 on frame-blended material is not going to produce good results.

If they really did transfer the film by using an unmodified movie projector, the script presented in the following link can sometimes improve the capture enough so that you can end up with something close to "frame accurate" and which can be modified with scripts like FILM9

The power of Avisynth - salvaging "botched" transfers of old 8mm films to DVD

It is a long post, and he doesn't describe specifically what he is doing until post #2 where he says:

This topic is about transferring old 8mm/Super 8 films to digital media, but only indirectly : it supposes that the transfer has already happened (in this case, to DVD format), but that it was so badly done as to be painful to look at (especially considering the price paid). This, unfortunately, seems to be a not uncommon occurrence.

[GillesH]

As John pointed out, FILM9 cannot work miracles and solve ALL problems.
If the source clips are of very poor quality, FILM9 cannot repair everything.

Occassionally it'll produce short bursts of static like this

However, in FILM9, the flickers are mainly treated during the 2nd Pass with "Stabilization" option activated.
To try...

[johnmeyer]

There are lots of ways to remove flicker. This includes the Deflicker plugin, which is an obvious choice, but also includes a "hidden" feature of MVTools2 which, if you use the correct DCT setting can do miracles with flicker. For tough cases, I use both.