InterFrame 2.8.2 - Framedoubling/60FPS/HFR conversion script

[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]

Originally Posted by edison

searchparam=1 works

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]

Originally Posted by 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.

You're very welcome, thanks for the support

Originally Posted by DeathTheSheep

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]

Originally Posted by DeathTheSheep

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.

[SubJunk]

Originally Posted by 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.

I've seen that effect before, too, in this video 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.

Originally Posted by edison

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:

Originally Posted by TempGaussMC_beta

# 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 ....



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.

[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]

Originally Posted by 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.

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]

Originally Posted by Shah Jahan

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"

Originally Posted by Shah Jahan

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.

Originally Posted by Shah Jahan

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.