VideoHelp Forum

Our website is made possible by displaying online advertisements to our visitors. Consider supporting us by disable your adblocker or Try ConvertXtoDVD and convert all your movies to DVD. Free trial ! :)
+ Reply to Thread
Page 11 of 11
FirstFirst ... 9 10 11
Results 301 to 323 of 323
Thread
  1. I wanted to thank you again for the support on my first undertaking of old film restauration. Especially the Film9-developers hint about the doubles in the video I got from the scan service was crucial but also all the code contributions of the pros that developed the parts of this software. As I am new to this I had no clue where to start. Meanwhile thanks to Film9 I have completed the two videos - I am quite satisfied with the results.

    I started here:
    https://streamable.com/vwapd

    and ended up here:
    https://www.youtube.com/watch?v=OulrI4yaz64

    https://www.youtube.com/watch?v=0_xbdR4qEnM

    (Workflow laid out in description)

    Film9 did a great job. I will carry the experience gained over to awaiting restoration projects. If there was a wishlist for Film9 I would write on it:

    More granular color control seperately on darks, midtones, highlight, r,g,b perhaps via gradation curves - in my case I had to to color correction with Color Mill in VD2 to achieve this

    If changing videos to edit in Film9 the VD2 window will jump to the foreground and covering the Film9 Window which has to be reselected to actually complete the desired task (this is minor and only usability)

    (Upscaling with superresolution - not quite sure if this is even really useful or feasable though)
    Quote Quote  
  2. Your results look really good.

    I too was quite aware of the lack of midtone correction. You have almost no shadow details in any scene, except near the end when the original exposure briefly increases and you can actually see the sides of the cabins.

    I do my gamma corrections in Vegas, using a histogram curve. There are also AVISynth tools for this.

    The color corrections were very nice. If you are doing those manually in Colormill, you definitely have the talent. The dirt removal was very good, and the grain reduction was absolutely spot on.

    The only other criticism is that the initial scans blew out the highlights. I've seen far worse, but a good transfer house should never allow this to happen. I do my own transfers and I always expose for the highlights, and then take care of the dark shadows in post. I use a video camera for capture, but the really high-end equipment (Cintel or Spirit) has a scanner that can capture the full range, so you don't have to do too much of this. Having said that, I had some client film rescanned by a Hollywood transfer outfit because it was going to be used by the Smithsonian channel. I had them send the transfers to me prior to sending to the production company and I had a chance to compare my results with theirs. They, of course, did a beautiful job on the highlights, but even with their equipment, I would have chosen to brighten the low end of the midtones.
    Quote Quote  
  3. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    Hi mcconnor,
    Thank you for your positive comments and advice.
    And a big cheer for your movie restorations.
    As we've always said, FILM9 is an intuitive interface to use Avisynth filters without directly using a script.
    Even though this work required a lot of study hours and tests, we can not thank enough Doom9 super developers and advisors.
    Without them, there could not be these kinds of results.
    We are following their studies and, perhaps, the future will take them into account for your recommendations.
    Cordially
    Quote Quote  
  4. Thanks for the kind words. That coming from you, Johnmeyer, is considered a high praise by me and encourages me to even improve next time.

    You are right. Details in the darks are the main problem here. I did the whole color grade with color Mill (besides the auto white from film9). One must note, that I always processed the whole 1h movie in one go. So there is no manual color grading for each scene. I took the video for from the first film9 run loaded it into vd2 and graded the color by correcting a difficult scene and then zapping through some others scenes to see if it fits. It went very well - I suppose because of the auto white. The darks remained a problem; in a dark scene I lifted the midtones just before they started to look unnatural. Then I lifted the darks and then took them down again just to the point where I started to see details vanishing {100% view in vd2). Increasing dark levels from there would only result in a muddy grey surface to appear with no detail in it so I finally concluded there was nothing in the raw anymore that could be salvaged. Essentially I was trying to spread the darks and darker midtones to a wider area as to expose detail that couldn't be seen in the raw scan. It worked to a fair degree but I hit the wall at some point. I also tried the gradation curves plugin for vd2 but I couldn't get better results and it was hard to control.

    But if you think after having seen the raw scan that there might be more to win and have a clue for me I might try again...
    Quote Quote  
  5. @johnmeyer and @GillesH got me trickled - It seems the problems of the (darker) midtones respectivly the lighter darks can be partly cured with the Auto-Gain feature of Film9. I am on it right now. Unfortunately this means applying stronger degrain is necessary becaus there is a lot of noise hidden in the darks which becomes prominent if brought to light... I will try to introduce more controlled grain later on to compensate. Color grading is changing of course - much less correction needed with Auto-Gain on.

    Observations / Questions:

    Frame interpolation in Film9 has a serious limitation (imho) - it does not take scene changes into account resulting in mixed images at every scene change. This results in flashing and if many scene changes occur this is irritating. Moreover it introduces issues with Motion compensation (e.g. Deshaker in VD2)! I had to cut out all scene changes with the adjoining frames. That is pretty much work if there are many scenes.... Isn't there an interpolation that can be integrated that does take scene changes into account? I found this: https://www.spirton.com/convert-videos-to-60fps/. It is based on avisynth and the sample videos look very good! Is even utilising the GPU... In the samples scene changes show no problem. FPS can freely be adjusted. Is there a possibility to review and perhaps use that in film9?

    Can the Auto-Gain be adjusted to e.g. in strength or limited to certain brightness regions? If so I probably would have tried to exclude the really dark darks from beeing ampliefied and thereby introducing a lot of noise....
    Quote Quote  
  6. Originally Posted by mcconnor View Post
    It seems the problems of the (darker) midtones respectivly the lighter darks can be partly cured with the Auto-Gain feature of Film9 ... Unfortunately this means applying stronger degrain is necessary becaus there is a lot of noise hidden in the darks which becomes prominent if brought to light...
    Try to be conscious of artifacts if you start trying to kill too much of that noise which appears when you gain the shadows. Most degraining algorithms get really funky when you try to kill shadow grain. You are better off leaving it rather than introducing grain reduction artifacts because the grain seems "natural" to most people's eyes, but the artifacts look awful.

    Originally Posted by mcconnor View Post
    Frame interpolation in Film9 has a serious limitation (imho) - it does not take scene changes into account resulting in mixed images at every scene change.
    I am not a fan of frame interpolation, even if you can solve the scene detection problem.

    You can certainly get scene detection to work if you use MVTools2 and set the scene detection parameters correctly. That part is actually quite easy. I don't know if this can be done within Film9 or whether it will have to be done as a separate operation in AVISynth, after you've done the other work (I don't use Film9 so I don't know what access you have to the internal algorithms).

    However, regardless of how you do the motion estimation, it breaks down far too often for the result to be watchable. I can refer you to dozens of posts over in doom9.org where we have explored the problems. As a result of those discussions, my transfer of a 1940s parade in Flint, MI has become a torture test clip that is used to illustrate all the pitfalls of motion estimation.

    Flint Michigan 1940 Parade

    Despite my dislike for the technology, I used "frame interpolation" for this clip because there is so much horizontal panning and, at 16 fps, such panning introduces pretty bad "judder" which is an unavoidable persistence of vision artifact (i.e., the artifact happens in your brain, not on the screen). This can be cured by adding the motion estimated frames, but you end up with all sorts of weirdness as objects get revealed in the background as the floats pass by; as vertical objects like shoulder-mounted rifles, flag poles move horizontally and, worst of all, moose antlers go passing by (1:42 mark). There are also artifacts on waving flays, people's legs "breaking" as they march, spokes on wheels that break and then reform, and so on.

    So, if you use the frame interpolation, be prepared that you are going to create these problems. There is no way around it. Even if you use thousand dollar motion estimation programs, you end up with the same problems.
    Last edited by johnmeyer; 23rd May 2019 at 12:19. Reason: added "1:42 mark"
    Quote Quote  
  7. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    The interpolator you are quoting is exactly the one used in FILM9.
    This is "InterFrame" from SubJunk. This is based on the SVPFlow study.
    Nothing is perfect in Interpolators. Have you tried the Duplicate function ?
    This is another way to multiply images. But, nothing is perfect with interpolator (as indicated by John).

    AutoGain is an "Automatic" function. There are no settings provided.
    Adding settings would require a lot of work.
    For the moment, I think that it is necessary to adjust then, either with the other settings of the window "Base Colors" or, then, with a NLE in post (as indicated by John).
    Quote Quote  
  8. Just so everyone knows, SVPFlow is indeed based on MVTools2, but it is subtly different. First of all, it does not contain all of the features of MVTools2 because it was originally produced for use as a real-time interpolator in order to increase the frame rate in real time, while you are watching the video, much like some higher-end TV sets do. To do this, it uses the display card's GPU and this processor sometimes produces different results than the computer's CPU.

    Second, I don't know if SVPFlow has kept up with all of the improvements in MVTools2. This is probably not too important for the use being discussed here, but I haven't followed the development of both branches of the original MVTools to know for sure. When in doubt, use one of the MVTools2 builds, although you'll have to consult some of the threads at doom9.org to know which one is currently preferred for this type of work.

    For super-critical work where I want to increase the frame rate in order to reduce or eliminate judder, but can't live with the artifacts, what I do is create two versions of the movie, one done with interpolation, and the other done with frame blending (frame duplication isn't an option when the goal is to increase the frame rate). Then, in my NLE, I play the interpolated version at slightly slower than normal speed and when I spot an artifact, I substitute the blended frame. For super-super critical work, I create a mask over the artifact and only use enough of the blended frame to hide the artifact. This is brutally time-intensive, so I've only done it for a few projects, and only when the play time is less than ten minutes.
    Quote Quote  
  9. Thanks for sharing those insights. To me it seems that either my source material is too taxing or that there is no scene change detection implemented /working.

    Scene 1
    Before Scenechange
    Image
    [Attachment 49130 - Click to enlarge]


    Interpolated 1/2 (Smooth)
    Image
    [Attachment 49131 - Click to enlarge]


    Interpolated 2/2 (Smooth)
    Image
    [Attachment 49132 - Click to enlarge]


    After Scenechange
    Image
    [Attachment 49133 - Click to enlarge]


    Scene 2
    Before Scenechange
    Image
    [Attachment 49134 - Click to enlarge]


    Interpolated (Smooth)
    Image
    [Attachment 49135 - Click to enlarge]


    After Scenechange
    Image
    [Attachment 49136 - Click to enlarge]


    This introduces issues with Autowhite (which is confused for some frames) resulting in bright flashing especially if the scene change is with high contrast and motion compensation (at least if "outsourced") which results in confusing sudden movement for some frames. All those frames have to be cut out at EVERY scene change (I checked my whole 1h film) which attributes to 3 minutes of playtime over all in my case and a lot of work.

    But at the same time the scenes itself have a very fluid interpolation which looks great and I couldnt spot so many artifacts so I wouldn't want to turn interpolation off or simply duplicate frames. Either it is just me or something is amiss with the interpolation at least concerning scene detection.

    As the examples of interpolation found on above linked site do not show any problem in scene changes I think it should be possible to somehow circumvent the issues I demonstrated. Most of the old movies contain a whole lot of scene changes. Cutting them all out is sometimes sacrifing critical content and always using up a lot of time. Sometimes you might want to cut out the scenes anyway (like in example 1 because of the dirt where the tapes are glued together). But most often scene changes are clean - Camera on/ camera off, no tape change) and normally wouldn't have to be cut. Cutting prior to processing with Film9 would avoid the problem but is cumbersome if there are many scenes. In my case well over 100 scenes - often quite short...
    Last edited by mcconnor; 24th May 2019 at 02:46.
    Quote Quote  
  10. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    The latest version of Interframe is from 2015 and has not been updated by its designer, but FILM9 uses the latest DLLs (SVPFlow 1 and 2).
    With Interframe, we do not use the GPU function because it is not compatible with all PCs.
    It is true that MVTools2 could be an alternative, especially that the updates are present (thanks to pinterf).
    It is also possible to use SVPFlow command lines without using Interframe.
    There are several new avenues to study. But it requires a lot of testing work with clips of different origins and nature ...

    Regarding the artifacts related to interpolation, we gave EXACTLY the same advice as John in the FAQs of our FILM9 website : (Film9/FAQ)
    Do 2 treatments with different methods (eg Interpolation and Duplication). Overlay the clips in the NLE and choose the best sequence.
    Of course, it takes a lot of time, but it limits or removes the artifacts.

    For changing scenes, MVTools2 has an MSCDetection function. But I have never tried. It is also to study ...
    Quote Quote  
  11. Originally Posted by GillesH View Post
    The latest version of Interframe is from 2015 and has not been updated by its designer, but FILM9 uses the latest DLLs (SVPFlow 1 and 2).
    With Interframe, we do not use the GPU function because it is not compatible with all PCs.

    <snip>

    For changing scenes, MVTools2 has an MSCDetection function. But I have never tried. It is also to study ...
    It is my understanding that Interframe is nothing more than a front end to SVPFlow and was created because the SVPFlow syntax is so strange (I've used it, so I know how difficult it is). However, it also hides quite a bit of the functionality in SVPFlow which, as I described previously, has already been stripped down from what is in MVTools2. Pretty much anything that is not part of changing frame rate (such as denoising) is not included in SVPFlow and therefore Interframe as well.

    As for the MSCDetection function in MVTools2, it is used to create a scene change flag for when, as an example, you want to automatically cut a video on scene boundaries. More relevant for this discussion are the several "thSCD" parameters which will reset the motion estimation at scene changes within MFlowFPS which is what you use to create synthesized frames. If you set this correctly, you won't get wild artifacts for the first few frames after a change. This will fix the problems that were asked about a few days ago.

    You can get access to these parameters in MVTools2, but not in SVPFlow or, by extension, Interframe, which is why I recommend MVTools2 for the interpolation.

    You can read more about these MVTools2 parameters here:

    http://avisynth.nl/index.php/MVTools#Common_parameters

    The Wiki says "MVTools" but it is for the later "MVTools2."
    Quote Quote  
  12. P.S. If you are up for it, here is a link to a tortiously long (28-page) thread in doom9.org where we discussed how to get better motion estimation using MVTools2, SVPFlow, Interframe, and other motion estimation tools.

    SVP-Like Frame Interpolation

    One person developed a script which attempted to find the artifacts and automatically mask them out. Pretty interesting stuff, although I never thought it worked well. It was in this thread that I first introduced my 1940 parade clip which has now become somewhat of a standard for testing motion estimated frame creation. You will find in this thread, and in several other places, that pretty much everyone agrees that the best results are still the ones from a script I created using MVTools2, and not from SVPFlow or any of the other alternatives. I'm not patting myself on the back. Instead, after dozens of hours of trying pretty much every setting available in MAnalyze, MFlowFPS, and the other relevant MVTools2 calls, I stumbled into a somewhat magic combination of calls that worked really well for low frame rate material (all of these tools work infinitely better if the source material is 60 fps instead of 16 fps because the smaller temporal gaps in 60 fps video are much easier to predict than at 16 fps). It even got made into a function by someone else (note the thSAD values):

    https://forum.doom9.org/showpost.php?p=1800439&postcount=3

    Having said that, the YouTube video I posted earlier was done with that script, and as you can see, it is loaded with artifacts. So while people seem to think my approach provided the best results, it still leaves a great deal to be desired.
    Last edited by johnmeyer; 24th May 2019 at 07:54. Reason: added another link
    Quote Quote  
  13. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    Originally Posted by johnmeyer View Post
    I'm not patting myself on the back.
    Thank you John for the advice. And no problem for your back. I know that your experience has always been good for everyone.

    Four or five years ago, I did comparative tests between Interframe and MVTools. At the time, Interframe seemed better.
    But since then, there have been developments on the side of MVTools2 and your advice deserve to review this old comparison.
    In the discussion you mentioned on Doom9, I had also tested FrameRateConverter without really convincing myself (among other things, too slow).
    So there is really a new study to carry out by integrating your jm_fps.avsi which is based on MVTools2 ...

    Because, in a script, it is also necessary to take into account the other associated filters and the reactivity that this will generate.
    There is a trade-off between quality and responsiveness. Some filters are slow, others use a lot of memory and PC resources, etc ...
    Quality is preferred, but sometimes we have bugs related to PC performance. Especially when the number of pixels to be processed is important (FullHD for example).

    We must therefore review all these new techniques that better manage the performance of a PC.
    There are all the latest DLL updates, but also, AVISYNTH-Plus !!!!
    Need to redo the kitchen and find new recipes!
    Quote Quote  
  14. [QUOTE=GillesH;2551394]
    Originally Posted by johnmeyer View Post
    There are all the latest DLL updates, but also, AVISYNTH-Plus !!!!
    Need to redo the kitchen and find new recipes!
    Me too. However, I am getting rather old and have a harder time updating stuff. However, I was forced into using AVISynth+ a few years ago in order to use a really good, new denoiser. I found the plus version to be remarkably better in every way. If you are using multi-threading (which I find mandatory in order to get any work done) your scripts will have to be slightly re-written, but after you've done it once, it doesn't take long for other scripts.

    I have not explored the 64-bit versions of AVISynth because, like almost everything 64-bit, there is so little improvement. My programming heritage goes all the way back to 1960s computers, and I went through the 4-bit, 8-bit, etc. progression. At those lower sizes, the bigger memory address space provided by more bits did indeed provide huge benefits. However, even with the bloatware created by modern, poorly-trained programmers, I have seen very little performance improvement in most apps when they get re-written for 64-bit. Having said that, when working with large megapixel RAW images in Lightroom, 64-bits is almost mandatory.

    So, a long way of saying that you should spend your time getting familiar with AVISynth+, but not the 64-bit versions. This is especially true in this case because there are still a lot of critical AVISynth DLLs that have not been re-written (it usually requires more than a simple re-compile) to take advantage of the 64-bit address space. Also, I don't think you are dealing with 4K video.
    Quote Quote  
  15. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    Originally Posted by johnmeyer View Post
    However, I was forced into using AVISynth+ a few years ago in order to use a really good, new denoiser.
    Please, what is this new Denoiser that you advise ?
    Quote Quote  
  16. Originally Posted by GillesH View Post
    Please, what is this new Denoiser that you advise ?
    KNLMeansCL.

    Here is the fantastically long (68 pages) thread in doom9.org where it is discussed:

    KNLMeansCL: OpenCL NLMeans de-noising algorithm [2018-01-29]

    Like all denoisers, it is not a panacea and can fail just as spectacularly as any other denoiser if used on video which has noise that is not what it is meant for.

    I got intrigued by it when I was not having any luck with other denoisers on some very dark 1980s VHS wedding footage that I was attempting to brighten which, of course, made all the shadow noise visible. This denoiser did a wonderful job with that.

    However, I have not tried it on any film scans so I don't know if it would work well to reduce grain noise. Also, it is horribly slow, and these various film restoration scripts are already pretty slow to begin with.

    P.S. Getting it up and running is a daunting task (it took me several days). You will find that the first twenty pages (which you can skim through quickly) are mostly about trying to get all the DLLs to load and also dealing with a lot of bugs back when it was first released.

    P.P.S. Here is a link to the point in the thread, after I got it working, where I answered VideoFred's question -- which was similar to yours -- about how well it worked:

    https://forum.doom9.org/showthread.php?p=1825149#post1825149
    Last edited by johnmeyer; 24th May 2019 at 14:21.
    Quote Quote  
  17. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    Thank you John for this info. Long reading, indeed.
    I even found that some confused "temperature" with "denoising" !!!!
    There would be pages to burn for those who prefer the heat ....

    Enough with the jokes. This KNLMeansCL filter seems more dedicated for use with GPU.
    Opinions are very divided. But the GPU and slowness are two incompatible ingredients for FILM9.
    So, we will continue with MVTools2 ...
    Thank you again and good luck for your work.
    Quote Quote  
  18. I did get it working and as I remember (and I may not be correct) I don't think I ever got my old NVidia card (10+ years old) to work with it, so I don't think it used GPU. However, I did get some of multi-threading to work.

    I was working on VHS, so it was SD resolution which obviously made it much faster. Also, it was a twenty minute wedding ceremony in a dark temple and that was the only part of the project that required this plugin. I think it finished overnight and I then got on to something else. If instead I had been transferring dozens of 400 foot film reels I would have been much more sensitive to the performance.

    Finally, it did such a good job on that particular noise that I tried it on other projects and the results were worse the MDegrain.
    Quote Quote  
  19. I wonder if there is some specific reason as to why it is not possible to select Cleaning Single/Double Action TOGETHER with Clean Dirts. In my experience the double action does seldomly improve on the single action but the clear dirt removes some stain and lighter dirt that is left over by the Single Action cleaning. So to benefit from both methods we have to run the file twice at the moment. Wouldn't it be possible to do that in one run (set both cleaning options active?)
    Quote Quote  
  20. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    In fact, we use 2 RemoveDirt functions (for darker or lighter tasks) with different settings.
    And these settings can not add up to the risk of causing slowdowns or bugs in the Avisynth calculation.
    But I know that John has a lot to work on this subject.
    Maybe, he can give information on a possible mix (black tasks and clear tasks) in a single function RemoveDirt ?
    Quote Quote  
  21. Hello, I am trying to restore with film9 some old 8mm from 70's. I got no error and image improves a lot, but the final result is that a sample 37 seconds of video results in a final project of 58seconds, a 10minutes in a 15 minutes, I have tried all the options i have seen and thought could solve the problem, but with no result, always in slow motion. I have w10 pro, film9 version 3.0, no errors or problems during intallation or
    restoration. The "speed" is set in 18fps as the original, I tried 16 and 24 and the same, slow motion. Do you have any idea what I am doing wrong?
    thank you
    Quote Quote  
  22. Member
    Join Date
    Dec 2014
    Location
    France
    Search PM
    efect2000,
    Have you tried the 2 functions in Project : WITHOUT Synchro and WITH Synchro.
    Depending on the capture mode, there may be time differences.

    What is your method of digitizing old 8mm ?
    Quote Quote  
  23. well, it's solved. I had in mind that the original was in 18fps and 8mm movie, now I selected video instead of old 8mm and now everything works perfectly. I use the old system like copying old cassete tepes: play-record. I project the movie on a piece of paper and I record it with the video camera, in fact it does a very good friend of me, that has a high quality video camara. The result is very very similar, if not the same, as a movie that a few years ago I took to a professional store, but now the machine does not work due to missing spare parts.
    Thank to all of you
    Quote Quote