InterFrame 2.8.2 - Framedoubling/60FPS/HFR conversion script

[PALpilot]

WOW thanks a lot! Add another Assumefps(25) and you have an awesome slowmotion script. It does so much better than the one on the mvtools2 page.
:thanks:

[SubJunk]

http://www.mediafire.com/file/33w33iqfjlgxa5l/svp_artifacts_supression_high_vs_interframe_720p.mkv

The artifacts supression settings of SVP is very useful in some case although somtimes will cause blending look, how can I enabled similar thing with InterFrame ?I finally got around to watching the video, great job on it. It's a great comparison.

WOW thanks a lot! Add another Assumefps(25) and you have an awesome slowmotion script. It does so much better than the one on the mvtools2 page.
:thanks:Awesome, thanks a lot for the feedback :)

[Dogway]

SubJunk: Just wanted to know if there is something to take care in mvtools for using GPU other than defining svp_flow_lib. Im using it just for mdegrain tasks, but I get artifacts in the few first frames of my encoded intermediary either using gpu or sse.

[SubJunk]

SubJunk: Just wanted to know if there is something to take care in mvtools for using GPU other than defining svp_flow_lib.No, there is no other way sorry.

[Dogway]

oops, I missed this post (http://forum.doom9.org/showthread.php?p=1491506#post1491506), looks like it only works for msmoothfps... thanks for the feedback

[CruNcher]

Most of the Hard work is still done on the CPU it wont be usefull for the hard core motion search part :)

[edison]

algo 13 vs algo 23 is almost same here:
http://www1.mediafire.com/imgbnc.php/3dfeef40760b0dd00ce1f73b4c0fa23d84a771c0c36d381638ceaec29fd945fa6g.jpg (http://www.mediafire.com/imageview.php?quickkey=t0fm0h5b5vfjh5j&thumb=4)

http://www.mediafire.com/file/guwggij7m89mhwz/multi_svp_nh_vs_origin.mkv

http://www.mediafire.com/file/w3ggw7dggd71l2t/multi_svp_nh_vs_interframe.mkv

http://www.mediafire.com/file/cvdupddtkkuu6gf/multi_interframe_algo_13_vs_default.mkv

origin 23.976p:
http://www.mediafire.com/file/z1zzg8qsnqen993/origin_23.976.mkv

[Didée]

algo 13 vs algo 23 is almost same here:
<pic>
"algo" changes the calculation of the interpolation.

The shown artifact is not a problem of interpolation ... the problem is the motion search has gathered bad motion vectors. Try tweaking MAnalyse settings. Lowering pzero, slightly raising pnew, maybe dct=5, search on a slightly blurred clip, .....

Remember, MVTools does not search "motion". It only searches for small squares that look similar to each other. ;)

[SubJunk]

Those errors can be avoided by using searchparam=1, but of course that has its own problems too

[Didée]

The search defaults are quite skewed, anyway. So yes, lets skew them even more. :D

[edison]

searchparam=1 works :)

http://www4.mediafire.com/imgbnc.php/ea54f29a7bac20b0fb4881060008e4b16f8c5b1cd98e5946d43d03d8eacf3d726g.jpg (http://www.mediafire.com/imageview.php?quickkey=wk2rjdduuynydd6&thumb=4)

[Didée]

But now it will miss much more motion that's different from global motion. :)

[edison]

maybe, but searchparam = 1 is better than searchparam = 2 in this case. I hope there is a better algorithm can handle this case .

[aegisofrime]

Subjunk, your modified MVTools is based on an older version, 2.5.11.1, correct? If so, would it be possible for you to release that under a different name? Soon QTGMC will require the latest MVTools, and thus using your MVTools effectively precludes me from using QTGMC, and vice versa.

[SubJunk]

Subjunk, your modified MVTools is based on an older version, 2.5.11.1, correct? If so, would it be possible for you to release that under a different name? Soon QTGMC will require the latest MVTools, and thus using your MVTools effectively precludes me from using QTGMC, and vice versa.Hi :) You can just rename the file

[Alek93j]

Hi SubJunk :) I read this in mvtools doc:
(Recent note: it is true for luma, but is not exactly corresponded to chroma pixels positions of internal MVTools interpolation. Nevertheless vectors and motion compensation are quite similar for usual clips, same chroma would be with src_left=0.5 for YUY2 and additionally src_top=0.5 for YV12).
I saw that you wrote this in the script
CustomPelClip = EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input), src_left=0.25)
I only partially understand your method; I tried doing it like this, shouldn't it be better?
FUNCTION pelclipfix(Clip input)
{
luma=EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input))
chroma=EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input),src_top=0.5,src_left=0.5)
MergeChroma (luma,chroma,1)
}

here the results:
your:
http://img225.imageshack.us/img225/8640/orig.png
my:
http://img839.imageshack.us/img839/9436/fix.png

does this make sense?

[Gavino]

luma=EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input))
Your change is correct for more accurate chroma, but you still need the src_left=0.25 for luma.

[Alek93j]

Is this related to avisynth's resize bug or is it an issue internal to mvtools?
Anyway:
FUNCTION pelclipfix(Clip input)
{
luma=EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input),src_left=0.25)
chroma=EEDI2(Input, field=1).Spline36Resize(2*width(Input), 2*height(Input),src_top=0.5,src_left=0.5)
MergeChroma (luma,chroma,1)
}

[Gavino]

Is this related to avisynth's resize bug or is it an issue internal to mvtools?
MVTools (see doc) requires the pelclip to have original pixels at (0,0) and at row and column positions with multiples of pel. EEDI2(field=1) does this when it doubles the height. However, the Avisynth resizers (by design, it's not a bug) preserve the image centre postion rather than the corner position, so to get the original pixels from the width doubling you need to specify src_left=0.25.

[Alek93j]

thank you for the answers

[SubJunk]

Cool :) I will have that fixed in the next version

[SubJunk]

1.11 is released, enjoy :)

[Alek93j]

I need to make a custom function with mvtools2 for slow motion needs in my editing works, but I noticed that your msmoothfps often works much better than I could manage, so I'd like to know what the finest, sadm, and sadgamma options in msmoothfps exactly do.

[SubJunk]

I need to make a custom function with mvtools2 for slow motion needs in my editing works, but I noticed that your msmoothfps often works much better than I could manage, so I'd like to know what the finest, sadm, and sadgamma options in msmoothfps exactly do.Mfinest creates a better pelclip than default and is only used in CPU mode.
Sadml is completely new to the custom mvtools and based on the ml from MMask, it defines a threshold at which vectors are marked as erroneous and then blurs/ghosts over the erroneous area using a different mask to ml. The lower the value, the more vectors are marked as erroneous. I recommend values between 150-250.
Sadgamma is the same as gamma in MMask.

[Ikaro]

Hello everyone, I use interframe from one month and I find it a great script to create interpolated frames in videos.
Thanks a lot for your work SubJunk!!!!!
But I need your help to solve this problem:

http://img231.imageshack.us/img231/6503/provasearchparam4.jpg

http://img808.imageshack.us/img808/2281/provasearchparam1.jpg

It's the same with ALGO 13.
This artifact is the biggest and most visible in the video, can you help me with the parameters??
TIA!!!

P.S. sorry for my bad, googled, english :-(

Max

[SubJunk]

Hi Ikaro. Sometimes that will just happen. mvtools gets confused because the sections of net are almost identical to eachother.
You can fix it by using blksize=32, but that will affect the overall smoothness.

[Rumbah]

Hi, I have a question about the modded mvtools.
I'd like to use Interframe with Dither. That uses a modded mvtools version itself. Is it possible to get both plugins working together or do I have to use an intermediate lossless video?

[SubJunk]

Sure, you can just rename the files however you want and include them when you want them. For example "mvtools-dither.dll" and "mvtools-interframe.dll".
Note that to do that they need to be in a directory other than the default one, because the default will autoload them.

[Rumbah]

Ok, thanks, I'm doing it in two steps now as loading both dlls at the same time does not seem to work for me, e.g.

dither()
Interframe()

In addition to that I notice that some settings are used here that are not mentioned in the documentation, like searchparam, algo and blocksize. Is there a document where I can read about them?

[SubJunk]

Yeah, using both at the same time wouldn't work. I thought you meant you wanted to use them separately.
Searchparam and blksize are the same as they are in the official mvtools documentation, and algo is basically an easy way of referencing mflow, mflowfps and mblockfps within the same function instead of separate ones.

[Rumbah]

Ok, thanks, then I've done it the right way.

Btw thanks for this script, I'm impressed with the result. When watching it frame by frame you'll notice errors but in motion it's great for the "normal" watcher.

[DeathTheSheep]

searchparam=1 works :)

http://www4.mediafire.com/imgbnc.php/ea54f29a7bac20b0fb4881060008e4b16f8c5b1cd98e5946d43d03d8eacf3d726g.jpg (http://www.mediafire.com/imageview.php?quickkey=wk2rjdduuynydd6&thumb=4)

Searchparam=2 is the best of all worlds. I just tried my script (see previous post) with your source with perfect results.

I will repeat this ad nauseam: for searchparam, anything higher than 2 is placebo, folly, or destructive, not helpful. I've shown many examples and others have, too. Make sure to change all references to "searchparam" to 2 in your script if they are higher.

[SubJunk]

Ok, thanks, then I've done it the right way.

Btw thanks for this script, I'm impressed with the result. When watching it frame by frame you'll notice errors but in motion it's great for the "normal" watcher.You're very welcome, thanks for the support :)

Searchparam=2 is the best of all worlds. I just tried my script (see previous post) with your source with perfect results.

I will repeat this ad nauseam: for searchparam, anything higher than 2 is placebo, folly, or destructive, not helpful. I've shown many examples and others have, too. Make sure to change all references to "searchparam" to 2 in your script if they are higher.For the record I still disagree ;)

[PALpilot]

Hi,

Again, superb tool. But I have another great use for it. I make HDR timelapse shots of landscapes from a tripod. I use my Canon 550D and the Magic Lantern application. I use an interval of 3-4 seconds and I use InterFrame to interpolate the in between frames so I get a movie.
So 1 HDR picture, 6-10 frames interpolated, next HDR picture. This results in a more alive landscape with small movements in trees. And InterFrame does a superb job interpolation cloudmovements so you can get a timelapse movie out of like 6 pictures.

I might stick with 1.10 sometimes because of the super smooth pans.

Note it doesn't work for some situations like running water of course.

[edison]

Searchparam=2 is the best of all worlds. I just tried my script (see previous post) with your source with perfect results.

I will repeat this ad nauseam: for searchparam, anything higher than 2 is placebo, folly, or destructive, not helpful. I've shown many examples and others have, too. Make sure to change all references to "searchparam" to 2 in your script if they are higher.

script of #136 not works.

http://www.gokuai.com/w/3nz240TD8CVl4i4X/deaththesheep_script_136.jpg

[SubJunk]

Hi,

Again, superb tool. But I have another great use for it. I make HDR timelapse shots of landscapes from a tripod. I use my Canon 550D and the Magic Lantern application. I use an interval of 3-4 seconds and I use InterFrame to interpolate the in between frames so I get a movie.
So 1 HDR picture, 6-10 frames interpolated, next HDR picture. This results in a more alive landscape with small movements in trees. And InterFrame does a superb job interpolation cloudmovements so you can get a timelapse movie out of like 6 pictures.

I might stick with 1.10 sometimes because of the super smooth pans.

Note it doesn't work for some situations like running water of course.I've seen that effect before, too, in this video (http://www.watchthinkchat.com/) for one. It looks great :)
Yeah it was a hard decision to trade smoothness for accuracy, really hard. I still question the decision sometimes, but oh well. Maybe I will change the smooth tuning to be like the regular one was in 1.10 so that users like you can continue to update.

script of #136 not works.Yes, for those types of errors usually changing searchparam to 2 will do nothing, only changing it to 1 will often fix it and even then sometimes not. The only sure fix is to use blksize 32 :) (like I said before)

[Didée]

Let's drop some hints.

In regard to those false-motion "jalousie window" artifacts shown on the last pages ...

Anyone ever spent a thought why this nasty artifact often occurs in areas of big local contrast (plus repetitive patterns, to make it worse).


A block with small local contrast typically has smaller SADs to blocks similar to it.

A block with big spatial contrast typically has bigger SADs to blocks similar to it.

Isn't it obvious that

"SAD_related_threshold = ONE_FIXED_VALUE_FOR_ALL_BLOCKS_NO_MATTER_THEIR_CHARACTERISTICS"

is going to give problems? The job is to identify motion. It doesn't matter if a moving block contains only weak detail, or very strong detail - here, "motion" is just a vector, and the vector by itself is not related to the block's content.

Still, the search engine will come to different decisions for either block type, because of the fixed thresholds. What is considered "good" for a soft block might be considered "bad" for a hard block. Or vice versa.

(The bell has been ringing for years now .... Basically I'm praying this since back in MVTools1 days)


Oh, one more hint:
# Create clip for motion search.
# Trick: weight-in a gaussian blur, to reduce # (Needed to later make MVDegrain work as expected ...the "SAD concept" doesn't fit
# local contrast of the searchclip. # to the realities -- so, let's tailor the realities to fit the darn concept ...)
This is not merely funny. It's one major key (out of two) why the whole shebang actually works .... :rolleyes:



Of course, you can continue to work around the problem by shrinking, shrinking, shrinking the actual motion search, like repeatedly suggested on the recent pages. Once you fully inhibit MVTools' motion search, you can be sure that it doesnt find "false motion" anymore. :p

[SubJunk]

I have never recommended shrinking motion search, it has only been a quick suggestion for specific videos. Actually, the only change I have made in the code has been to increase the default searchparam from 3 to 4 ;)
As always, if you have an improvement to the script I'm happy to try it.

[Emulgator]

Avisynth open failure:
Script error: there is no function named "MFinest"
(InterFrame.avsi, line 132)
(InterFrame.avsi, line 50)

Ah. mvtools 2.5.11.6 is needed. My fault. I was on 2.5.11.2 16bit.

[SubJunk]

Glad you figured it :)

[Emulgator]

many thanks, SubJunk ! Interframe works quite good to get some framerepeat-flagged AVCHDLite 1280x720x25p to more fluid 1280x720x50p.

A smallish suggestion: Working in varying OSes it may become tedious to hand edit any plugin paths when swapping machines.

A global variable helps here:

#To automate varying plugin call paths a script named AviSynthPluginsDir.avsi has to reside in AviSynths plugins directory.
#It uses the call GetWorkingDir() which requires GetSystemEnv.dll by stickboy to be in AviSynths plugins folder.
#AviSynthPluginsDir.avsi contains only one line: global AviSynthPluginsDir = GetWorkingDir()


Now any script may call its respective plugins, even in separate folders.

Version for the "single big bag of plugins":

InterFrame(FlowPath=AviSynthPluginsDir)

Version for separated plugin folders where a folder named "InterFramePlugins" shall contain the plugins to be loaded exclusively for InterFrame:

InterFrame(FlowPath=AviSynthPluginsDir + "InterFramePlugins/")

Unfortunately I don't seem to get a 64-bit GetSystemEnv.dll. Did anybody find or build one ?

[SubJunk]

Hi Emulgator, I'm glad you like it, thanks for the feedback :)
With regards to the suggestion, it seems like that is something that would be better left to the user, rather than me including it. Some people like to have their plugins in different directories from eachother that are nowhere near the default one so I think the current system is more flexible for everyone :)

[SubJunk]

1.12 is released, enjoy :)

I should also note that I was experiencing stability problems with SET's recent AviSynth 2.6 update (2011.07.19) with most scripts, so if you use his builds and have been getting crashes I recommend going back to his 2009.09.19 build

[Yellow_]

I'm enjoying using your Interframe script, but wonder if there is any way to minimise the warping effect around movement, same problem with Twixtor I believe, any sort of masking that can be done?

[SubJunk]

There are developments happening with the SVP developers right now that improve that by making the forward and backward vector information synchronised instead of separate.
It isn't released publicly yet, though. When it is I will update InterFrame for it :)

[SubJunk]

1.12.1 is released :)

[Shah Jahan]

I'm using your Guide to framedoubling/60FPS conversion and at moving objects it creates a hazy vapor like thing at moving objects and I'm not sure if I'm using the Placebo preset. How to select the preset for encoding? I know how to for real time, but the guide for encoding doesn't mention on how to do it for encoding? I'd want to use the Placebo preset, if it's not already. Is there any way to make it work with MediaCoder?

And when you update your realtime versions, then do you also update the encoding versions or do we have to do it manually? If yes, then how?

Thank you.

[SubJunk]

The guide download and the InterFrame download are updated at the same time.
The placebo tuning won't offer big improvements, it is only slightly better.
The hazy vapor will still be there on placebo tuning. There are quality improvements coming soon.

[Shah Jahan]

The guide download and the InterFrame download are updated at the same time.
The placebo tuning won't offer big improvements, it is only slightly better.
The hazy vapor will still be there on placebo tuning. There are quality improvements coming soon.
Thanks for the reply.
So what are the current settings used in encoding? (Placebo, medium...?)
Can you please make different encoding settings (like it is for realtime encoding) and post it on your website OR kindly give me the placebo settings for encoding?
Can you please give any tentative date about the next version which has some quality improvements? Or atleast allude if it's within days/weeks/months?

Thank you.

[SubJunk]

Thanks for the reply.
So what are the current settings used in encoding? (Placebo, medium...?)By default it's Medium and you can change it to whatever you like by specifying "Preset"

Can you please make different encoding settings (like it is for realtime encoding) and post it on your website OR kindly give me the placebo settings for encoding?I'm not sure what you mean. The script is always the same whether realtime or not. Please see the documentation, it is linked to from the first post and is in the download.

Can you please give any tentative date about the next version which has some quality improvements? Or atleast allude if it's within days/weeks/months?I guess weeks.