InterFrame 2.8.2 - Framedoubling/60FPS/HFR conversion script

[Stormborec]

Can it work for single core too?

[SubJunk]

@Stormborec I think it can work for single core, just use Cores=1.

@TCMullet that reminds me, I think your script should use the same amount of cores for SVP and SetMTMode. It's probably not related to this issue but might make things more stable.

[TCmullet]

Originally Posted by SubJunk

@TCMullet that reminds me, I think your script should use the same amount of cores for SVP and SetMTMode. It's probably not related to this issue but might make things more stable.

I'm very green at SetMTmode, and am using only because I was told to. Here's that tail code, plus I have another question somewhat related.

ConvertToYV12()
SetMTMode(2)
InterFrame(Cores=2,Tuning="Film")
SR(1280,720,2)

As you can see, I had already set cores equal as you suggested (after the fact). What I'm asking about now relates to SR in addition.

I had set InterFrame cores to 1 and SR to 2, think that SR (SuperRes 2.0, which you're probably familiar with) needs more (a lot more) cpu than InterFrame. But if I set both to 2 and I only have a 4-core system (and no hyperthreads), would I then be choking out the OS (Windows 7)?? Seems to me that Windows would be choked off and the whole system would bogg down, knowing that the OS DOES need cpu cycles.

I'd like your thoughts. In the meantime, I did some tests, and I surprisingly (but maybe not to you) found that 2 and 2 ran a okay and maybe even a little better than 1 and 2. A possible explanation might be that (if true) all application tasks have lower priority than any OS task, and therefore the OS never gets short-changed.

So what is the general formula for when you have 2 multi-threading apps running on a 4-core system? (Do I need to give a core to the system?)

[SubJunk]

In general I'd say to let every program/app/script use max cores. As long as everything is normal and healthy on the computer (like it's not being overclocked too much) it should be fine. Like you said, process priorities make a difference.

I hadn't heard of SuperRes but I'll give it a go now, I have an upscaling toolchain I regularly use so I might be able to add it to that. Actually I got the Blu-ray for a movie the other day and compared it to a DVD upscale I did of the same movie, and my upscale was better quality, which is a real shame.

[TCmullet]

Originally Posted by SubJunk

In general I'd say to let every program/app/script use max cores.

Please explain what you mean by "max cores", using my example of a 4-core system (no hyperthreads).

How many threads should be allocated to InterFrame and how many to SuperRes, given that they are both in the same script? If you answer that, I'll know which way to interpret your comment.

[SubJunk]

AviSynth isn't great at multithreading so it's impossible to tell what will work best here, but I would try 4 for everything at first and see how it goes. A lot of times multithreading in AviSynth is a process of trial and error and gives different, unpredictable results for different computers, along with SetMemoryMax.

[TCmullet]

Originally Posted by SubJunk

AviSynth isn't great at multithreading so it's impossible to tell what will work best here, but I would try 4 for everything at first and see how it goes. A lot of times multithreading in AviSynth is a process of trial and error and gives different, unpredictable results for different computers, along with SetMemoryMax.

Oh so 2+2 is not valid logic. Okay. At present, I'm 16 hrs into a 21 hr run, then another VDub job kicks in for probably 17 hrs more. Then I can try 4+4 for the next task. However, I have a Cpu usage icon that continually shows bar graph for all 4 cores. All 4 are floating near the top, so I feel I'm doin pretty good w/2+2.

But then in task mgr, it adds up to a total, which fluctuates from 85-94%, adding the unused portions of individual cores' usage together. So maybe with 4+4 the total will get closer to 100%, not a trivial jump.

Your prior ambiguity caused me to think. AviSynth is so amazing to me and it boggles my mind to even try to think about how it can even work. I've been assuming that InterFrame and SR are concurrent processes, that is, both running "at the same time". In a 1-core system, obviously the def. of concurrency helps us to see that time slicing lets both run but only one at a time. Knowing now that I have 4 cores, I was thinking of the need to allocate them out. But maybe InterFrame can use 4 cores for a bit (to create the new frame), then SR can use 4 cores for doing ITS task on that frame, and so each step takes it's turn using all resources while its up to bat.

Hey, I'm using the unknown (to me) SetMTmodes. My statements are:
SetMTMode(3,4)
SetMTMode(2)

If I try your idea of 4+4, what would I set those 3 mode values to?

[SubJunk]

With those values, you're already using 4 cores; the second argument in SetMTMode is the number of cores to use.

[TCmullet]

Then you're saying to leave the SetMTs alone. Great.

[TCmullet]

Originally Posted by Atak_Snajpera

Does anybody know how to increase motion search range or precision in order to eliminate "double leg effect"?

BTW. Source is 1080@25fps. I'm trying to achieve nice looking slow motion effect.

I skimmed pages 16 (in 2012) to the present (p. 37) again to look for any responses to Atak's question, and if the answer or even any significant discussion was there, I missed it.

Either it can be done (and what settings would accomplish it) or it cannot be done (and we need to be told that explicity or we--some of us--will be eternally wondering "what am I missing").

His photo doesn't show the problem as explicitly as does a current short clip of mine. Here's a link to a 21 frame Lag. file: (It was too big to attach here, so I'm hosting it at my own server.)
http://www.tomsgoodfiles.com/Tom's-t...97.2015-02.avi

Here's a script:

SetMemoryMax(512)
SetMTMode(3,4)
AviSource("D:/Video-Work/Tom's-test-clip2997.2015-02.avi"   )
ConvertToYV12()
SetMTMode(2)
InterFrame(Cores=2)

Clip is 21 frames long, and after running the script, frame 11 is probably the best one to show the problem. 10 and 12 show the original leg motion, with 11 being interpolated.

Subjunk will you please download my video and run this script (fudging path as needed), look at those 3 frames (10-12) and please explain whether InterFrame can be user-tweaked to prevent frame 11 (the leg portion) from looking like a blending of frames 10 and 12? And if not, could you please say so and (hopefully) explain why InterFrame can do "ball-left + ball-right ==> ball-middle" but cannot do "leg-left + leg-right ==> leg-middle"?

And if not, can you please theorize "out loud" what would have to be done to make it accomplish this?

(I may post the 3 stills involved, later if I can.)

[TCmullet]

Here are frames 10, 11, and 12 in .jpg. 11 is the interpolation between frames 10 and 12.

(3 pics are worth at least a few dozens words, eh?)

[SubJunk]

Using GPU=true increases the quality significantly, including with that clip.

[TCmullet]

Originally Posted by SubJunk

Using GPU=true increases the quality significantly, including with that clip.

But that doesn't make sense. I thought GPU was there only as a means to get extra cpu power in addition to the cores in the main computer, so as to speed up the conversion a bit (like having one additional core). Why would borrowing the processor in the graphics card have anything to do with the logic of Interframe beyond that? And why should it?? (Seems to me it should NOT.)

Any way, I tried:
InterFrame(Cores=2,GPU=true)

and it bombs immediately with a VirtualDub message:
Avisynth open failure:
SVSmoothFps: unable to load library given in 'compose' [c:\program files (x86)\AviSynth 2.5\plugins\svpflow_gpu.dll]
(InterFrame2.avsi, line 156)
(InterFrame2.avsi, line 58)

(then line 6 in my script)

So at the moment I'm not able to see what "great improvement" you are alleging.

[StainlessS]

SVPFlow_GPU.dll requires OpenCL.dll, perhaps you dont have it. (I think it comes with graphics card, so perhaps yours [like mine] is incompatible).

[TCmullet]

Originally Posted by StainlessS

SVPFlow_GPU.dll requires OpenCL.dll, perhaps you dont have it. (I think it comes with graphics card, so perhaps yours [like mine] is incompatible).

My system has an OpenCL.dll file in the folder, C:\Program Files (x86)\SVP\ even though I don't have a separate graphics card beyond the builtin Intel adapter.

I await SubJunk's full analysis of all this. I am "disappointedly" eager to scrap days of processing I've already done, and start over correctly.

[TCmullet]

While I am continuing to await SubJunk's full analysis of this problem, I searched on 3 systems and found that my video system (the i5 system w/quad-core--way the fastest system I have and on which I was intending to do all video processing) DOES have the file, as I already said. I also found it in C:\windows\system32 on another system (XP), a tiny Atom 4-core system.

So I have experimentally copied the file from my SVP folder to C:\windows\sysWOW64 and the error message that now comes up is a little different:

SVSmoothFps: unable to init FlowLib [code 0x10000]
instead of
SVSmoothFps: unable to load library given in 'compose'

(I'm not sure why I have an SVP folder; probably leftover from long-ago attempts.)

[Groucho2004]

Originally Posted by TCmullet

So I have experimentally copied the file from my SVP folder to C:\windows\sysWOW64 and the error message that now comes up is a little different:

OpenCl.dll usually comes with the driver of a specific GPU. Copying it from one system to another is a really bad idea.

[TCmullet]

Originally Posted by Groucho2004

OpenCl.dll usually comes with the driver of a specific GPU. Copying it from one system to another is a really bad idea.

Thanks, Groucho. I'm just trying to redeem time and attempt to go forward a bit, while waiting for SubJunk to reply to all of this.

[Groucho2004]

Originally Posted by TCmullet

My system has an OpenCL.dll file in the folder, C:\Program Files (x86)\SVP\

Neither SVP nor the graphics driver install would put that file there.
Edit: I take that back, SVP does install OpenCL.dll.

[TCmullet]

My adapter is "Intel HD Graphics". A friend suggested I google. I found Intel's
opencl_runtime_15.1_x64_setup.msi
which alleges to install both 32 and 64 bit versions of openCL.dll for MY builtin Intel graphics adapter.
I installed and rebooted.
I now have OpenCL.dll dated 01/25/2015 in both directories system32 and sysWOW64
Similar error message, but a little different:
SVSmoothFps: unable to init FlowLib [code 0x100FF]

Am greatly hoping we can get OpenCL and InterFrame "GPU=true" to run on this my fastest machine. (Unless the functionality that SubJunk briefly said was needed via GPU=true could be implemented via GPU=false.)