MCTemporalDenoise [v1.4.20 - Update 2010/07/02]

[totya]

Hi,

I think I found all needed filters (with exact version), except this:
Deblock_QED (25.may.2010)

This version date is 2010-10-16. This is good or not?

Second question: Of many needed filters they are newer version, these better than (recommended) olders, or not?
Example latest MVTools version is 2.5.11.3 etc...

Thanks!

[the_weirdo]

Originally Posted by totya

Hi,

I think I found all needed filters (with exact version), except this:
Deblock_QED (25.may.2010)

This version date is 2010-10-16. This is good or not?

Second question: Of many needed filters they are newer version, these better than (recommended) olders, or not?
Example latest MVTools version is 2.5.11.3 etc...

Thanks!

IMHO, you should use latest version of those needed filters, except RemoveGrain + Repair.

[totya]

Originally Posted by the_weirdo

IMHO, you should use latest version of those needed filters, except RemoveGrain + Repair.

Hi,

thx, but recommended version of RemoveGrain + Repair already latest: v1.0PR, this is mean I think: "v1.0 pre-release - latest edition with SSE3 fix"

[the_weirdo]

Originally Posted by totya

Hi,

thx, but recommended version of RemoveGrain + Repair already latest: v1.0PR, this is mean I think: "v1.0 pre-release - latest edition with SSE3 fix"

You're right. I've misremembered

[jmac698]

It's sad that no one has noticed the installer
http://www.sendspace.com/file/1gggec

[totya]

Originally Posted by jmac698

It's sad that no one has noticed the installer
http://www.sendspace.com/file/1gggec

Hi, I don't see this link on first post... and anywhere.

Symtomps:

- Default instal path is wrong, if avisynth installed outside on default path (good path from registry)
- Deblock_QED_MT2.avsi is not recommended version too
- mt_masktools-25.dll is not recommended version (latest), and recommended avisynth-MT verison is 2.60 see, and this version need mt_masktools-26.dll

But nice pack

[jmac698]

Thanks for the feedback! (finally)
I had mentioned it here:
http://forum.doom9.org/showthread.php?t=162622
It's also here:
http://avisynth.org/mediawiki/MCTemp...quired_Filters
But this is a good thread, it just wasn't a top thread lately so I didn't know about it. I guess people are not finding information, but at least check the avisynth wiki as there's some good information there.

- Default instal path is wrong, if avisynth installed outside on default path (good path from registry)

That's why I made it customizable. To do it properly, I have to make a more advanced installer which can read the registry.
I should warn you that I put a few files in system32, fft3w.dll (required) and avsrecursion (often used, but not required by this filter). Ideally I want to make an installer for many things "filter pack".
You can open the .exe in 7-zip and see everything.

- Deblock_QED_MT2.avsi is not recommended version too

Can you point me to the correct file?

- mt_masktools-25.dll is not recommended version (latest), and recommended avisynth-MT verison is 2.60 see, and this version need mt_masktools-26.dll

I didn't install any avisynth-mt. mt_masktools-25.dll *should* be from a48. I can also include mt_masktools-26.dll, there's no problem there. I only use 2.58 myself, haven't used mt yet

[asarian]

Originally Posted by asarian

Actually, had the same issue the other week, with Der Untergang. At first (and, in hindsight, on earlier occassions) I thought MCTemporalDenoise was simply hanging; but it's not: it simply drops to like 0.08 fp/s. This was on a full 1080p source, uncropped, and with 'medium' for settings. Switching to 'low' made the problem go away, but that was obviously not what I wanted.

I experienced the issue again, soon thereafter, with The Shadow, cropped to 1040p, and also on 'medium' settings.

I eventually 'solved' the matter by simply splitting the film up vertically into 2x 540p parts (each with a 16px overlap), and processing each with MCTemporalDenoise individually, only to splice them back together later on. Not ideal, of course.

Speaking of feedback, does anyone else experience the same issue? Or is no one else processing HD material?

[jmac698]

Ok, the installer version is correct masktools, latest version

58440c2964e3b8a3899d265b644d05bc *mt_masktools-25.dll

[totya]

Originally Posted by jmac698

Can you point me to the correct file?

No, this is my already qestion.

Originally Posted by jmac698

I didn't install any avisynth-mt. mt_masktools-25.dll *should* be from a48

But recommended version is MaskTools (v2.0a43).

I can open your installer with 7z, and I see, fftw3.dll is very outdated. New version.

Thanks, and I see wiki too...

[jmac698]

Ok,
The original Deblock_QED_MT2.avsi is at http://www.64k.it/andres/dettaglio.php?sez=avisynth. This version requires masktools <a35 to work, because of new "bias" parameter added. The newer Deblock_QED_MT2.avsi made some minor changes, and now needs masktools a45+ to work. Using a48 should be ok. The requirements are actually different if you update some of the scripts.
So just tell me, does my pack work as is?

Yes I used old version fftw3, once I tried to use new version and it didn't work for me. I should test again.

Update:
I checked the changelog from a43-a48. Some bugs were fixed and new features added, but nothing incompatible.

Alpha 48 :

added : signed and unsigned binary shifts to LUTs, ceil/floor/trunc float -> int conversions. @ is now equivalent to °

Alpha 47 :

added : signed and unsigned binary operators to LUTs

Alpha 46 :

fixed : with mt_lutf, "std" mode wasn't working. Could also affect "std" with mt_luts and mt_lutsx

Alpha 45 :

fixed : mt_clamp on sse2 platforms with resolution not multiple of 64
added : mode parameter to mt_lutspa, to clean up biased/relative mess

Alpha 44 :

added : new mode (weighted sum) for mt_luts and mt_lutsx

I will also list the differences in the new Deblock_QED_MT2.avsi:

# Changes 2008-08-18: (Didée)
# - Replaced the ugly stackXXX cascade with mt_LutSpa() (requires MaskTools v2.0a35)
# - Changed Quant and Offset defaults to 24,28,2,4,4,8

# Changes 2010-05-25:
# - Explicitly specified parameters of mt_LutSpa()
#   (required due to position of new 'biased' parameter, starting from MaskTools 2.0a43)
# - Non mod 16 input is now padded with borders internally

# Changes 2010-08-18:
# - Replaced AddBorders with PointResize
# - Changed Quant and Offset defaults to 18,19,3,4,1,1 to reduce blurring

# Changes 2010-10-16:
# - Replaced 'relative' with the new 'mode' parameter in mt_LutSpa(), starting from MaskTools 2.0a45
# - Changed Quant and Offset defaults to 24,26,1,1,2,2 to increase effectiveness, but still within sensible limits.
#   (see for details: http://forum.doom9.org/showthread.php?p=810932#post810932)

[totya]

Originally Posted by jmac698

Ok,
The original Deblock_QED_MT2.avsi is at http://www.64k.it/andres/dettaglio.php?sez=avisynth. This version requires masktools <a35 to work, because of new "bias" parameter added. The newer Deblock_QED_MT2.avsi made some minor changes, and now needs masktools a45+ to work. Using a48 should be ok.

Thanks for the info!

Originally Posted by jmac698

So just tell me, does my pack work as is?

Not much different for me "packs"

Originally Posted by jmac698

Yes I used old version fftw3, once I tried to use new version and it didn't work for me. I should test again.

All new version is work's for me (need libfftw3f-3.dll... rename it to fftw3.dll. I read this somewhere on doom9 forum). You can run benchmark "benchf.exe -opatient 64 128 256 512 1024 2048 4096" and you see speed different.

Thanks!

[jmac698]

Ok, I will update the one file and try to read the correct installation directory.
Thanks for the testing and feedback.

[Boulder]

Originally Posted by asarian

Speaking of feedback, does anyone else experience the same issue? Or is no one else processing HD material?

I've noticed a similar issue sometimes with SD material as well. Using a radius of over 2 frames is somewhat slower (or at least was with my old Q6600), and that's the big difference between medium and the lower defaults.

[asarian]

Originally Posted by Boulder

I've noticed a similar issue sometimes with SD material as well. Using a radius of over 2 frames is somewhat slower (or at least was with my old Q6600), and that's the big difference between medium and the lower defaults.

The problem for me is not so much that 'medium' takes a bit longer than 'low' (which is totally reasonable), but that sometimes 'medium' (or higher) slows down by an order of magnitude (like I said, ca. 0.08 fp/s on my i7 980X). In fact, it slows down so much that you think the process is stuck; only it isn't (it looks correctly when you examine the output) -- it just goes extremely slow; like where it would suddenly take 700 hours to complete.

Like I stated, the slowdown doesn't always happen; merely on some sources (I suspect because of the the grain being pretty intense).

[cretindesalpes]

When the framerate suddenly drops by one order of magnitude or even more, try to increase the frame cache memory, for example with SetMemoryMax(1024) at the beginning of your script.

[asarian]

Originally Posted by cretindesalpes

When the framerate suddenly drops by one order of magnitude or even more, try to increase the frame cache memory, for example with SetMemoryMax(1024) at the beginning of your script.

Thanks! Never really understood the memory max variable. Gonna give it a try (when the current render is done); and I'll let you know if it worked.

[LaTo]

It would be great to find a script & sample that isolate the problem. Someone?

[chainring]

Originally Posted by asarian

Speaking of feedback, does anyone else experience the same issue? Or is no one else processing HD material?

I'm currently doing some 1080 BD to 720p downconversions and have no problems. If I try to do anything with SetMTMode, no matter the mode and number of threads, all hell breaks loose.

i7 2600K
4 GB RAM
Win7 x64 Ult

Avisynth 2.6 with SET's latest MT .dll
Latest MeGUI set for dev updates
All the correct versions of Avisynth plugin's for MCTD, except I'm using the latest Masktools.

[chainring]

Originally Posted by cretindesalpes

When the framerate suddenly drops by one order of magnitude or even more, try to increase the frame cache memory, for example with SetMemoryMax(1024) at the beginning of your script.

I noticed that with, IIRC, when playing with SetMTMode.