Multiple troubles editing video from 8mm film; suggestions welcome

[Groami Geistalt]

Hi

Some of the video I'm trying to clean-up has multiple problems and I'm not sure what I'm looking at.
(I apologise if this file is too large, I'd normally upload something half this size or less, but in this case the details might help.)

I'm using AVSPmod to clean transferred 8mm film image.
I'm new to the work so I'm adding Filters to my Script as I gain a reasonable understanding of each.
Even so, some of the Filters keep their default settings because I don't understand enough to change them.
Until now these settings have worked well 'out of the box' but now I'm looking at some frames from near the beginning of a film spool, which are in such a mess that I'm not sure which problems have priority--problems whose treatment might help fix some of the other symptoms I'm seeing.

Here are the Filters I'm using thus far. (The settings seen here haven't been changed from the last job I did.)

ConvertToYV12()
Tweak(bright=-95, cont=1, sat=1 )
# WarpSharp(depth = 30, threshold = 40 )
TTempSmooth( maxr = 7)
TTempSmooth( lthresh = 4, cthresh = 5 )
TTempSmooth( lmdiff = 2, cmdiff = 3 )
TTempSmooth( strength = 2 )
TTempSmooth( scthresh = 12.0 )
TTempSmooth( fp = true )
TTempSmooth( vis_blur = 0)
WarpSharp(depth = 30, threshold = 40 )
XSharpen( strength = 30, threshold = 40)
MC_Spuds(frames = 2, strength = 4, search=2, blocksize=8, pel=2, edgeclean=true, temporal=false, aggressive=true)
fps=23.976

If you have the time to check this list of Filters and share your comments, I'm interested in your opinion about how to shuffle these Filters so that they're working in an order that best treats the problems, and your opinions on how to tweak settings.
Should any of these Filters be #'d out this time round, and something else used here?

I've played with the settings a bit already but brighten/darken didn't set the stage for other improvements, and I'm not sure how to interpret the cracks I'm seeing: are they 'noise' or 'artefacts'? Is MVAnalyse over-analysing or under-analysing... I don't know what to think.

Here's a good example of the video stream, the whole thing is about twenty seconds long, but I'm hoping to learn how to interpret problems here, and learn how best to treat them, as well as learn when to give up trying to tweak the image, owing to the limits of the quality of the original film.

Thanks for your attention and any help.

[newpball]

If would be nice if you could upload a short pre and post video sequence.

[Groami Geistalt]

If would be nice if you could upload a short pre and post video sequence.

Hi
Will this do? I'm not sure how much or how little to upload. (For that matter it's the first time I've tried to upload a bit of video too.)

[poisondeathray]

It's extremely unlikely that you'd want to call TTempsmooth so many times

Explain how this transfer was done.

What are your "original" files ?

What did you upload? It's uncompressed RGB and there are clean borders around it that were not part of the original film and probably not in the original transfer

(In the future, if you upload uncompressed material, you can zip it up to reduce the filesize . Or use lossless video compression)

[jagabo]

Also, upload video samples (every frame) not a severely decimated clip.

[LMotlow]

Your sample has been deinterlaced and converted to uncompressed RGB24. I doubt that it was captured as RGB24, but was probably captured as YV12 DV-AVI, interlaced. Was the original 8mm film shot at 10fps? I thought 8mm movie film was shot at 18 or 20fps. Movie film is progressive -- are you deinterlacing because the film was recorded from a projection screen with a DV camera and interlaced media? Was it recorded onto VHS tape and then captured to DV? Did you want your final video output to be RGB? What do you intend to do with the final video? View it on TV with a DVD or BluRay player? Send DVD copies to friends? Should we make more guesses, or could you furnish a little more information?

You have a journey ahead with Avisynth and VirtualDub, but everyone has to start somewhere so don't feel alone. Your script uses two old-fashioned avisynth ports of some old VirtualDub sharpeners. Did you intend to use two sharpeners? I think you might have felt that you needed to do so because your script ran the TTempSmooth softener/smoother filter repeatedly 7 times, 6 of the runs using settings that were already defaults.

These 7 lines from your script:

TTempSmooth( maxr = 7)
TTempSmooth( lthresh = 4, cthresh = 5 )
TTempSmooth( lmdiff = 2, cmdiff = 3 )
TTempSmooth( strength = 2 )
TTempSmooth( scthresh = 12.0 )
TTempSmooth( fp = true )
TTempSmooth( vis_blur = 0)

are the same thing as:

TTempSmooth(maxr=7)
TTempSmooth()
TTempSmooth()
TTempSmooth()
TTempSmooth()
TTempSmooth()
TTempSmooth()

Perhaps what you meant to do was 1 filter instead of 7 repetitions (?) If so, one statement is all you would need:

TTempSmooth(maxr=7)

Running this statement:

Tweak(bright=-95, cont=1, sat=1 )

with two of the parameters at their defaults, is the same thing as running this:

Tweak(bright=-95)

I noticed that you darkened a video that already looks rather dark in your sample, smoothed the video 7 times, sharpened it twice, then deinterlaced it. TTempSmooth is a time=based temporal smoother designed for progressive video -- it is less effective with non-progressive sources. Apparently your original capture had many duplicate frames -- if that was the case, you would have to take different measures to use temporal cleaners effectively with frames that repeat themselves. Temporal cleaners are based on the way an image changes over time -- if the image doesn't change for two or three consecutive frames, the filter doesn't "see" much noise.

Did you really want to change the fps of 8mm film to 23.976? Shouldn't it be around 20fps? Have you eve heard of adding pulldown (telecine) to get 20fps film up to 29.976?

Let us know if your sample was unfiltered or not. It's the wrong colorspace, so we're not sure what was done to the original capture.

You can submit more than 2.1 seconds of original, unprocessed capture video. If you read the format list in the forum download window, you'll see allowed file sizes. You don't have to hit the maximum size allowed, but 2.1 seconds of already-processed video is not very much for analysis. If you don't know how to cut a clean unprocessed sample, let us know. 8 to 10 seconds with motion should be enough.

[Groami Geistalt]

Thank you all. I'll digest this, try to fix what needs fixing, and then reply.

[Groami Geistalt]

It's extremely unlikely that you'd want to call TTempsmooth so many times
Explain how this transfer was done.
What are your "original" files ?
What did you upload? It's uncompressed RGB and there are clean borders around it that were not part of the original film and probably not in the original transfer
(In the future, if you upload uncompressed material, you can zip it up to reduce the filesize . Or use lossless video compression)

1) (I'll start using TTempsmooth as described further on in the replies)
2) The transfer was done with a Retro-8 Film Transfer Unit.
3) The original files are MJPEGs which I stitched into an .AVI using VirtualDub.
4) I uploaded a section of the .AVI initial made by VirtualDub from those MJPEGs.
5) (I'll try to remember to .zip any other video I upload.)

Thanks!

[poisondeathray]

Individual MJPEG's ? 1 MJPEG per frame ? or why would you need to "stitch" them ? Why didn't you upload an original clip ?

When you "stitch" together in vdub, you need to use video=>direct stream copy. You didn't upload MJPEG, you uploaded uncompressed RGB

You don't need to "zip" files up all the time; only ones that are uncompressed. MJPEG should already be compressed so it would be unnecessary to zip. If you zip up your file that you uploaded early, you will notice the filesize about 6MB, < 1/3 of the size

[newpball]

1. Does the scanner have an infrared channel?
2. What was the capture resolution, 8mm has a higher resolution than SD video.

With respect to filters, I would start with color correction, aging film tends to get darker and colors shift. Then see if you can remove some grain. I do not see the point in sharpening unless you are going to view this on a small screen perhaps.

[Groami Geistalt]

Hello

>Your sample has been deinterlaced and converted to uncompressed RGB24.
>I doubt that it was captured as RGB24, but was probably captured as YV12 DV-AVI, interlaced.

The Retro-8 Film Transfer Unit created the MotionJPEGs, and the first thing I did was stitch them together in VirtualDub.
Then I called that .AVI with AVSPmod, and one of the jobs the Script did was convert the color to YV12

> Was the original 8mm film shot at 10fps? I thought 8mm movie film was shot at 18 or 20fps.

So did I. I haven't been clear on why the Film Transfer Unit captures at 10fps, and I've been trying to understand what I've been reading about changing FPS. ("Pull-down" has been referenced often but until now I thought VirtualDub's 'change frame rate' function was all I needed to do.) My approach thus far has been to clean the image with the Script and then change the frame rate at the end, when VirtualDub creates the final .AVI file.

> Your script uses two old-fashioned avisynth ports of some old VirtualDub sharpeners. Did you intend to use two sharpeners?
> I think you might have felt that you needed to do so because your script ran the TTempSmooth softener/smoother filter
> repeatedly 7 times, 6 of the runs using settings that were already defaults.
> ... one statement is all you would need: TTempSmooth(maxr=7)

I probably copy-pasted that from somewhere, felt it was 'probably working right', and waited until advice came along to show me how to tidy-up the syntax.
I'm experimenting with the sharpeners. Sometimes one seems to work better than the other, sometimes both together seem to do the best job. I REM-out all of them and then test them one at a time.

> Tweak(bright=-95, cont=1, sat=1 ) is the same thing as running Tweak(bright=-95)

Thanks.

> I noticed that you darkened a video that already looks rather dark in your sample,
> smoothed the video 7 times, sharpened it twice, then deinterlaced it.

That brightness setting was from a previous job, I hadn't tried to do anything with the video yet. I just listed the Filters I'd be using.
The sample I sent was from a section of video VirtualDub had created from the MJPEGs before anything was done to the frames.

> TTempSmooth is a time=based temporal smoother designed for progressive video --
> it is less effective with non-progressive sources.

Now I recall reading about 'progressive' and 'non-progressive' but it didn't sink in and I haven't been thinking about it. I'll make a special note to re-read about this in more detail and in terms of temporal smoothing.

> Apparently your original capture had many duplicate frames --
> if that was the case, you would have to take different measures to use temporal cleaners effectively with frames that repeat
> themselves. Temporal cleaners are based on the way an image changes over time -- if the image doesn't change for two or
> three consecutive frames, the filter doesn't "see" much noise.

As far as I know the Retro-8 Film Transfer unit just made a single MJPEG from each frame of film.
If my sample is suggesting otherwise then maybe it's due to the nature of the AVI I made when I stitched together the MJPEGs.
Apparently I should find a better way of working with those.
Maybe I should load the MJPEGs directly into VirtualDub and try to Preview changes made in the Script that way, then save one .AVI at the end. I think I'm messing things up making that initial .AVI and then trying to run the Script on it. It might be another one of my unecessary steps...

> Did you really want to change the fps of 8mm film to 23.976? Shouldn't it be around 20fps?
> Have you eve heard of adding pulldown (telecine) to get 20fps film up to 29.976?

I heard of it, but not properly. I was using Frame Rate conversion some other way, I wasn't understanding clearly what to do.

> You can submit more than 2.1 seconds of original, unprocessed capture video.
> If you read the format list in the forum download window, you'll see allowed file sizes.
> ... 2.1 seconds of already-processed video is not very much for analysis.
> If you don't know how to cut a clean unprocessed sample, let us know.

Thanks very much!

[Groami Geistalt]

(It's getting late here... Thank you, all, for the replies. I'll resume this sometime Wednesday. Your questions are introducing things to me, I'll try to get clear on what's being said. Thanks again.)

[Cornucopia]

The Retro8 is a fine device, but it does seem to have some flaws, in that:
1. It doesn't make use of an infrared channel
2. It has max resolution of 1280x720 (I'm assuming due to type of sensor used)
3. It has only a few output options with it's capture software (which one must use for it to work): MJPEG video, JPEG pic sequence, BMP pic sequence.
4. It doesn't seem to have professional WB or exposure controls, in fact few controls at all.
5. It's dynamic range is limited.
6. It's noise reduction capability is limited (but at least it doesn't seem to exhibit undue smoothing).
7. It defines a standard framerate as 17Fps.

Even given all that, it seems to put out above average pictures.

I think a lot can be done to improve the workflow by just:
1. Using BMP sequence (at least you might be getting 4:4:4 uncompressed then)
2. Using AssumeFPS() of the proper rate (whatever that should have been with your recording)
3. No colorspace conversion, unless necessary - then, as few as possible (1?)
4. Maintaining uncompressed (or using Losslessly-compressed) through to the end of edit stage

More could be done, but one would need to use different/improved components for that to happen.

Scott

[LMotlow]

Hmmm....Sorry, I made an incorrect assumption. Many users make a film transfer by using a VHS or digital camera to record a film while it plays on a projection screen, but that method has endless problems. I have some learning to do myself about your MJPEG device. Thanks for the information.

[jagabo]

4) I uploaded a section of the .AVI initial made by VirtualDub from those MJPEGs.

Upload the original MJPEG file(s) the capture device created.

[MovieStuff]

The Retro8 is a fine device,

Thanks!

but it does seem to have some flaws, in that:
1. It doesn't make use of an infrared channel

Slide scanners employ two different light sources when using Digital Ice or other automated dirt/scratch removal tech. A beauty pass is made on the slide using regular light and then a second pass is made using infrared. There really isn't an infrared channel as much as it is just two images created with different wavelengths of light. The infrared will only be blocked by scratches or opaque dirt and that is used to create a map of where the imperfections are located. But this can only be done on a stationary frame where the registration of the image does not change between the beauty pass and the infrared pass. To have the Retro unit stop on each frame for a couple of seconds to accommodate the dual exposures would easily mean an entire day per 400 feet of film. While technically possible, I would never be able to stay in business selling such a slow scanner. As it stands, the older Retro units ran at 2fps which meant 4 hours per 400 feet and there was enough gripe about that for incentive to increase the speed up to 15fps in our new Retro Pro units. So what you see as a "flaw" is really just trying to meet the demands of the actual market rather than the select few that are not trying to meet customer deadlines.

2. It has max resolution of 1280x720 (I'm assuming due to type of sensor used)

That is correct. We tested early on making a 1080p unit. However, research showed that the most popular HD flat screen sold worldwide was 42 inches or smaller. At that size, the visual difference between 720 and 1080 for 8mm film is practically imperceptible and not enough to warrant the impact on pricing, speed and marketability. As it stands, the Retro units can work with most any off the shelf laptop. Moving up to 1080 would mean a more expensive, custom built desk top PC to meet the demands of being able to handle frame by frame scanning at the higher data rate. Also, due to the difference in scanner head for 1080, the price of the Retro unit would have more than tripled. So, again, while technically possible it would severely limit the market base just for the bragging rights that we produced a 1080p scanner. And, since fewer people would be buying them, that would raise the price of the unit even more because I have to maintain the same overhead regardless of whether I put a 720 scanner head or a 1080 scanner head in the Retro unit.

3. It has only a few output options with it's capture software (which one must use for it to work): MJPEG video, JPEG pic sequence, BMP pic sequence.

Actually, you have many options. The unit scans totally codec free. We do not scan to any type of video file. Each frame of film is scanned to an independent digital still frame. They can be compressed as high resolution JPEGs or uncompressed as BMPs. The software will allow you to play back that sequence immediately to check your capture without having to render anything. If you want to use the captured footage in your NLE, then you can export as an HD or SD video file in the MJPEG codec as a .MOV file or you can export as a numbered image sequence in SD or HD as compressed high resolution JPEGS or you can export as uncompressed PNG, BMP or TIFF files. The native HD resolution is 720 but the software will upscale seamlessly to 1080 for NLEs that prefer 1080 files.

4. It doesn't seem to have professional WB or exposure controls, in fact few controls at all.

Actually, you have many professional controls. Rather than use a primitive white balance function, the software actually has full control over all RGB channels as well as full gamma control. Exposure control is achieved by simply turning a knob on the Retro scanner. The combination of the exposure knob and the gamma control gives you full contrast control. Not sure what else you would want in a scanner of this price.

5. It's dynamic range is limited.

Actually, the 24 bit ( 8:8:8 ) dynamic range is so wide that most people complain about the scans looking too flat.

6. It's noise reduction capability is limited.

You are correct! In fact, the unit's noise reduction is so limited as to be non existent. There is grain reduction which is, of course, different than noise reduction. All digital cameras (whether still or video) use a fairly massive sharpening algorithm to make sense of the captured signal. Turn off that sharpening function on even the most expensive camera and the picture is totally fuzzy and out of focus. Turn on the sharpening function and you often see over enhancement on the edges because real time video cameras have to cut corners when sharpening due to the speed in which the frames must be processed. We chose the Chameleon camera because it was one of the few that had no sharpening function and we developed our own sharpening algorithm based specifically on the demand of scanning grainy 8mm movies. Our grain reduction is passive rather than active. When you activate it, you are really just turning down the degree of sharpening applied to the image. We felt this was a better approach than applying active grain reduction knowing that some people might be bringing these images into Neat video or other post apps to do their own sharpening and grain reduction. If we applied active grain reduction, then you could end up with over processing of the image.

7. It defines a standard framerate as 17Fps.

Actually, we do not. The standard frame rate for regular 8mm film is ostensibly 16fps. The standard frame rate for super 8 is ostensibly 18fps. I use the term "ostensibly" because experience has shown that most cameras do not run dependably anywhere close to this speeds. We offer the 17fps rate as a convenience on exported .MOV files because it is comfortably between 16 and 18fps and looks perfectly normal. However, the RetroScan software will also output 1:1 .MOV files with no speed applied. Anyone wanting a more exact frame rate of their choosing can bring those 1:1 files into their NLE and apply any frame rate they wish, which is easy to do. People concerned with the highest quality will no doubt be using the numbered image sequences which are by their nature 1:1 anyway, so you can have uncompressed quality at any frame rate you wish.

Even given all that, it seems to put out above average pictures.

The Academy of Motion Picture Film Archives agrees with you.

I think a lot can be done to improve the workflow by just:
1. Using BMP sequence (at least you might be getting 4:4:4 uncompressed then)
2. Using AssumeFPS() of the proper rate (whatever that should have been with your recording)
3. No colorspace conversion, unless necessary - then, as few as possible (1?)
4. Maintaining uncompressed (or using Losslessly-compressed) through to the end of edit stage

See all of the above to address this list.

My best,
Roger Evans
owner, MovieStuff, LLC
http://www.moviestuff.tv

[newpball]

Slide scanners employ two different light sources when using Digital Ice or other automated dirt/scratch removal tech. A beauty pass is made on the slide using regular light and then a second pass is made using infrared. There really isn't an infrared channel as much as it is just two images created with different wavelengths of light. The infrared will only be blocked by scratches or opaque dirt and that is used to create a map of where the imperfections are located.

Doesn't TIFF support an infrared channel?

To have the Retro unit stop on each frame for a couple of seconds to accommodate the dual exposures would easily mean an entire day per 400 feet of film. While technically possible, I would never be able to stay in business selling such a slow scanner. As it stands, the older Retro units ran at 2fps which meant 4 hours per 400 feet and there was enough gripe about that for incentive to increase the speed up to 15fps in our new Retro Pro units. So what you see as a "flaw" is really just trying to meet the demands of the actual market rather than the select few that are not trying to meet customer deadlines.

Ideally the customer should have the option.

Does a customers who just found his long lost immensely valuable one of a kind 8mm movie from one of his ancestors really care if it takes twice as long if the quality can be improved?

[MovieStuff]

Slide scanners employ two different light sources when using Digital Ice or other automated dirt/scratch removal tech. A beauty pass is made on the slide using regular light and then a second pass is made using infrared. There really isn't an infrared channel as much as it is just two images created with different wavelengths of light. The infrared will only be blocked by scratches or opaque dirt and that is used to create a map of where the imperfections are located.

Doesn't TIFF support an infrared channel?

Yes, TIFF can accommodate an infrared channel but that is different than saying the unit has no infrared channel. I was attempting to clarify how the process of dual exposures actually worked. While possible to scan as TIFFs, they are quite large and scanning at 15fps would be problematic for the average customer unless they had a very powerful, custom built PC. Slowing the unit down for a couple of seconds per frame to accommodate infrared scanning would take a very long time, as already indicated.

To have the Retro unit stop on each frame for a couple of seconds to accommodate the dual exposures would easily mean an entire day per 400 feet of film. While technically possible, I would never be able to stay in business selling such a slow scanner. As it stands, the older Retro units ran at 2fps which meant 4 hours per 400 feet and there was enough gripe about that for incentive to increase the speed up to 15fps in our new Retro Pro units. So what you see as a "flaw" is really just trying to meet the demands of the actual market rather than the select few that are not trying to meet customer deadlines.

Ideally the customer should have the option.

But they already do. There is software that will do post processing scratch and dirt removal without the need to increase the complexity and price of the Retro units to satisfy the few that want to do that kind of work

The thing to consider is there is a difference between "film migration" and "film restoration". If people could project their old 8mm films as safely and as conveniently as sticking a DVD in a player, then that is exactly what they would do and there would be little need for film transfer services or scanners, in general. And if they were projecting those old 8mm films, they would be seeing them with all the artifacts they've gotten used to, including scratches, faded color, lint, etc. And they would be satisfied with that because it cost them nothing. But they can't project their films safely or easily so "film migration" is the primary goal of most people that simply want a safe and convenient way to view what they already have. "Film restoration" is an additional service that attempts to remove the scratches and other artifacts that came with age and improper storage and viewing of the films. While the Retro units have some powerful image manipulation tools, they are really there to help make the film look on digital as close to the original as possible since reversal film was never meant to be duplicated, especially to something with as limited a pallet as digital video. There are some very powerful image restoration softwares out there. Whether the Retro units have an infrared mapping capability or not is really incidental to the amount of time needed to process and clean up 28,000+ frames of film per 400 foot roll of film. Adding infrared capability makes the units more expensive than necessary for the majority that don't need it just to satisfy the few that already have other options without it.

Does a customers who just found his long lost immensely valuable one of a kind 8mm movie from one of his ancestors really care if it takes twice as long if the quality can be improved?

Which "customer" are you referring to? The business owner that buys the Retro unit and has to meet a monthly budget or their customer that wants film transferred? Who pays for the additional price? How much can a business owner reasonably charge for spending 24+ hours on a single roll of film? How much is an individual willing to pay for a Retro unit to transfer their home movies? What if they don't need scratch removal and don't want to pay more for a unit than necessary?

I agree that if you already have the technology and it doesn't cost you anything but time and you can afford that time then, sure, why not? But the reality is that most people buying the Retro units are business owners. If they want to provide restoration services, there's software on the market that will allow them to do that if they have customers willing to pay a premium. If not, then the business owner hasn't paid for capabilities in the Retro unit he doesn't need.

It isn't always about what's technically possible because everything is possible. But the R&D for that possibility is going to be reflected in the price of the unit, so there's a practical side to producing units like this.

Roger Evans
owner, MovieStuff, LLC
http://www.moviestuff.tv