CCD - Camcorder Color Denoise (Chroma spatial denoiser) - AVISynth+ version - Page 2

pinterf · 15th June 2020, 14:42

I'm not sure but it seems that the filter is processing (width*height) bytes of input, instead of (pitch*height) bytes?

Avisynth internally works with 64-byte alignment, and provides buffers as such. The structures passed to the filter contain pitch and even a modulo (pitch-rowwidth) number.

Artifacts on top of RGB32 means that the last X bytes are left unprocessed (packed RGB is top-bottom)

StainlessS · 15th June 2020, 14:54

It does though work in x64 Vdub2 [EDIT: direct load of AVI]

EDIT: I've given Shekh a poke, maybe he enlightens us soon.

EDIT: I've also tried with AVs script

Code:

AviSource("D:\Parade.avi")
ConvertToRGB32

VDub2 x64 and still no problem.

EDIT:

Quote:

I'm not sure but it seems that the filter is processing (width*height) bytes of input, instead of (pitch*height) bytes?

More specific than prev given snippet, so above statement looks correct.

Code:

# FIX VDubFilter.dll Weird BUG, but not the x64 bug
# XMOD=16 YMOD=1   # OK
XMOD=1 YMOD=16 # BAD, Prob in XMOD

Crop(0,0,Width/XMOD*XMOD,height/YMOD*YMOD)

pinterf · 15th June 2020, 15:23

It's just a feeling, I see that depending on the horizontal crop (starting from mod16, then crop(0,0,-1,0) .. crop(0,0,-15,0) filter is giving bigger and bigger unprocessed area at the frame end (visible on top).
Note that -1 is 4 bytes, and a horizontal crop of -15 is 60 bytes.
VDub filter can hint that it requires a buffer with a specific minimum alignment, which request is surely fulfilled by avisynth by having 64 byte aligned NewVideoFrame.
Can you test it with avs 2.6 and width -1, -2 and -3 crop values (it has 16 bytes alignment)?

StainlessS · 15th June 2020, 15:38

Quote:

Can you test it with avs 2.6 and width -1, -2 and -3 crop values (it has 16 bytes alignment)?

Clarify, Avisynth 2.60 standard x64 ?

pinterf · 15th June 2020, 15:42

32 bit. I can see the problem with 32 bit version as well. No difference between 32 or 64 bit versions with my AviSynth+, both have black (total and partially filled) lines.

StainlessS · 15th June 2020, 15:51

Hmmm, 32 bit looks ok here.

This is weird [AVS std 2.60 x86]

Code:

# 64 bit VDF fails with error 0xC1, but filter works OK in VDUB2 x64, possible AVS+ VDubFilter.dll problem.

# VDub GUI "Show Noise" and "Disable" args not available external to VDUB, ie not usable in Avisynth.
# VDF accepts only RGB32
Function VD_CCD(clip clip, int "threshold", bool "multithreading",bool "ShowNoise",Bool "Amp",Int "preroll") {
    threshold      = Default(threshold,30)                                      # Default 30 as in Vdub
    multithreading = Default(multithreading,true)                               # Default true
    ShowNoise      = Default(ShowNoise,False)                                   # Simulated
    Amp            = Default(Amp,True)                                          # Amplify if ShowNoise
    preroll        = Default(preroll,1)                                         # We've given default of 1 frames pre-rolled, maybe only 0 needed.
    Assert(clip.IsRGB32,"VD_CCD: Need RGB32")
    LoadVirtualdubPlugin("D:\VDUB\ccd_32bit.vdf", "_VD_CCD", preroll)           # Change Path to VDF
#   LoadVirtualdubPlugin("D:\VDUB\ccd_64bit.vdf", "_VD_CCD", preroll)           # Change Path to VDF   ## Getting Error OXC1 for 64 bit VDF plugin
    result = clip._VD_CCD(threshold,multithreading?1:0)
    return (!ShowNoise) ? result : (!amp) ? result.Subtract(clip) : result.Subtract(clip).Levels(126-8,1.0,126+8,0,255,coring=false)
}

THRESHOLD=30
SHOWNOISE=false
AMP      =false

AviSource("D:\parade.avi")

ConvertToRGB32(matrix="rec601")

# FIX VDubFilter.dll Weird BUG, but not the x64 bug
#XMOD=16 YMOD=1 # OK
#XMOD=1 YMOD=16 # BAD
#Crop(0,0,Width/XMOD*XMOD,height/YMOD*YMOD)

# Here, 488x360

Crop(0,0,-1,0)  # FINAL result still shows 488x360 ???  [EDIT Oops, 488 not 448]


VD_CCD(threshold=THRESHOLD,shownoise=SHOWNOISE,amp=AMP)

info

I have to leave very soon, got to hit the shop before it shuts.

EDIT: If Switch on SHOWNOISE, then fails at subtract due to different size frames.

EDIT: Avs+ v3.61 test 8 x86 returns width 487 as it should, so looks like Avs 2.60 std has bug in VDFilter.

shekh · 16th June 2020, 08:56

Quote:

Originally Posted by StainlessS

EDIT: I've given Shekh a poke, maybe he enlightens us soon.

Not sure what you expect, surely a filter can have bugs. We dont know if x64 version was built from same source

StainlessS · 16th June 2020, 10:20

OK Shekh, was hoping that you might have some magic words to say, thanks for taking a look.

Quote:

Clarify, Avisynth 2.60 standard x64 ?

I guess I got myself a bit confused there, thats what happens when you are bout 12 hours past your bedtime and are attempting to do bout 17 things at once.
I had to go back and read thread from start to figure out what the problem was/problems were.
The 64 bit problem that I am having is that I cannot load x64 plugin, with error 0xC1.
x86 problem is the weird stuff at top of frame.
Also, I had gotten it into my head that VDFilter.dll calls VDub/VDub2 when it just has some kind of emulation layer for VD.

Anyways, had some sleep now and hopefully have un-confused myself, a bit.

pinterf · 16th June 2020, 10:40

Anyway, without seeing the filter's source code it's waste of our time.

StainlessS · 16th June 2020, 10:51

Also found out why I had problems with x64, I was trying to load x64 plug into x86 app,

Anyways, tried

Code:

THRESHOLD=30
SHOWNOISE=false
AMP      =false

AviSource("D:\parade.avi")
ConvertToRGB32(matrix="rec601")
# Here, 488x360
Crop(0,0,-1,0)    # Now 487x360
VD_CCD(threshold=THRESHOLD,shownoise=SHOWNOISE,amp=AMP)
info  # FINAL result still shows 488x360 ???

With avs standard v2.58, v2.60, v2.61, and all of them have problem in VDFilter showing width as 488 when we have cropped 1 pixel from RHS, should show 487.

Quote:

Anyway, without seeing the filter's source code it's waste of our time.

OK.

pinterf · 16th June 2020, 11:05

Quote:

Originally Posted by StainlessS

With avs standard v2.58, v2.60, v2.61, and all of them have problem in VDFilter showing width as 488 when we have cropped 1 pixel from RHS, should show 487.

If I remember well from my quick debug session, this filter reports that it supports v14 interface, while classic Avisynth only supports up to v6 (?). But I don't know what different behaviour it means.
Can you load Avisynth+'s vdubfilter dll into classic Avisynth?

StainlessS · 16th June 2020, 11:31

Quote:

Can you load Avisynth+'s vdubfilter dll into classic Avisynth?

v2.58, "Not a v2.5 avisynth filter" [or words to that effect].
v2.60,2.61, Seems to Work OK, prev error script returns 487x360 as it should. [and also same rubbish at top of frame for the bugged vdf filter].
So in v2.60/2.61 standard, LoadPlugin on Avs+ VDFilter.dll, Installs AVS+ LoadVirtualDubPlugin() and fixes the erroneous 488x360 bug [produces correct 487x360], and seems to work just fine.

zybex · 17th June 2020, 08:50

I talked with the author of the filter, informed him that we were going to create a remake of this filter under AVISynth+, and asked him to share the filter’s algorithm and / or sources, he said that he would send me some specific information in a couple of days when he returned from the trip.

zybex · 18th June 2020, 09:04

Here are the filter sources:

[URL deleted]

The author of this filter Sergey Stolyarevsky kindly provided them to us.

And he promised to join our discussion on this topic in a few days.

==================
Updated fixed version here:
https://forum.doom9.org/showthread.p...25#post1916925

pinterf · 18th June 2020, 10:24

Quote:

Originally Posted by zybex

Here are the filter sources:

https://mega.nz/file/uxxXCYwT#CthrV9...SbFJhPTbuDsNw8

The author of this filter Sergey Stolyarevsky kindly provided them to us.

And he promised to join our discussion on this topic in a few days.

Good. After having a quick look into the source I noticed that pixels are addressed as (y*width + x) and not (y * pitch + x) which explains why passing a buffer with an (64 byte) aligned pitch will not be processed entirely. So - if I'm right - Sergey can fix it quickly.

zybex · 27th June 2020, 19:09

Updated fixed version CCD V1.9 (Sources + Binaries)

https://mega.nz/file/GsYlQayQ#LO7qq8...KuQRCamtUe0lO4

StainlessS · 27th June 2020, 20:05

Cheers Sergey Stolyarevsky and Zybex.

Sergey 1400 · 1st July 2020, 08:39

Hello! I am Sergey Stolyarevsky, the author of the filter. Version 1.9 solved the problems?

zybex · 1st July 2020, 09:51

Quote:

Originally Posted by Sergey 1400

Version 1.9 solved the problems?

Hi, Sergey.
Glad to see you on the forum.

pinterf now on vacation until about July 11:
https://forum.doom9.org/showthread.p...64#post1916864

StainlessS · 1st July 2020, 15:50

Hello and welcome to the forum Sergey 1400

15th June 2020, 14:42	#21 \| Link
pinterf Registered User Join Date: Jan 2014 Posts: 2,314	I'm not sure but it seems that the filter is processing (widthheight) bytes of input, instead of (pitchheight) bytes? Avisynth internally works with 64-byte alignment, and provides buffers as such. The structures passed to the filter contain pitch and even a modulo (pitch-rowwidth) number. Artifacts on top of RGB32 means that the last X bytes are left unprocessed (packed RGB is top-bottom) __________________ AviSynth+ on github, Other repos: RgTools, Masktools2, MvTools2, TIVTC, Average

15th June 2020, 15:23	#23 \| Link
pinterf Registered User Join Date: Jan 2014 Posts: 2,314	It's just a feeling, I see that depending on the horizontal crop (starting from mod16, then crop(0,0,-1,0) .. crop(0,0,-15,0) filter is giving bigger and bigger unprocessed area at the frame end (visible on top). Note that -1 is 4 bytes, and a horizontal crop of -15 is 60 bytes. VDub filter can hint that it requires a buffer with a specific minimum alignment, which request is surely fulfilled by avisynth by having 64 byte aligned NewVideoFrame. Can you test it with avs 2.6 and width -1, -2 and -3 crop values (it has 16 bytes alignment)? __________________ AviSynth+ on github, Other repos: RgTools, Masktools2, MvTools2, TIVTC, Average

15th June 2020, 15:42	#25 \| Link
pinterf Registered User Join Date: Jan 2014 Posts: 2,314	32 bit. I can see the problem with 32 bit version as well. No difference between 32 or 64 bit versions with my AviSynth+, both have black (total and partially filled) lines. __________________ AviSynth+ on github, Other repos: RgTools, Masktools2, MvTools2, TIVTC, Average

16th June 2020, 10:40	#29 \| Link
pinterf Registered User Join Date: Jan 2014 Posts: 2,314	Anyway, without seeing the filter's source code it's waste of our time. __________________ AviSynth+ on github, Other repos: RgTools, Masktools2, MvTools2, TIVTC, Average

18th June 2020, 09:04	#34 \| Link
zybex Registered User Join Date: Apr 2007 Posts: 33	Here are the filter sources: [URL deleted] The author of this filter Sergey Stolyarevsky kindly provided them to us. And he promised to join our discussion on this topic in a few days. ================== Updated fixed version here: https://forum.doom9.org/showthread.p...25#post1916925 Last edited by zybex; 27th June 2020 at 19:10.

17th June 2020, 08:50	#33 \| Link
zybex Registered User Join Date: Apr 2007 Posts: 33	I talked with the author of the filter, informed him that we were going to create a remake of this filter under AVISynth+, and asked him to share the filter’s algorithm and / or sources, he said that he would send me some specific information in a couple of days when he returned from the trip.

27th June 2020, 19:09	#36 \| Link
zybex Registered User Join Date: Apr 2007 Posts: 33	Updated fixed version CCD V1.9 (Sources + Binaries) https://mega.nz/file/GsYlQayQ#LO7qq8...KuQRCamtUe0lO4

27th June 2020, 20:05	#37 \| Link
StainlessS HeartlessS Usurer Join Date: Dec 2009 Location: Over the rainbow Posts: 10,980	Cheers Sergey Stolyarevsky and Zybex. __________________ I sometimes post sober. StainlessS@MediaFire ::: AND/OR ::: StainlessS@SendSpace "Some infinities are bigger than other infinities", but how many of them are infinitely bigger ???

1st July 2020, 08:39	#38 \| Link
Sergey 1400 Registered User Join Date: Jun 2020 Posts: 1	Hello! I am Sergey Stolyarevsky, the author of the filter. Version 1.9 solved the problems?

1st July 2020, 15:50	#40 \| Link
StainlessS HeartlessS Usurer Join Date: Dec 2009 Location: Over the rainbow Posts: 10,980	Hello and welcome to the forum Sergey 1400 __________________ I sometimes post sober. StainlessS@MediaFire ::: AND/OR ::: StainlessS@SendSpace "Some infinities are bigger than other infinities", but how many of them are infinitely bigger ???