Simple MDegrain Mod v3.1.2d - A Quality Denoising Solution

[Auliya90]

@Dogway: Glad to see the new update here thank you ,i like your Smdegrain so much.

[Auliya90]

I being trying new update v3.1.2 i been facing some difficulties to make prefilter 4 works but after reading about it i found a i need an function call knlmeansCL for that now prefilter has gone threw but one more error poped up i m still struggeling to find out what is Mfilter and how it can be used if i want to . I tried Mfilter=true but seems like MFilter doesn't work like this i get this error "the name arguement "mfilter" to SmDgrain had the wrong type" .Please Dogway can u tell how MFilter could be used, thank you.
Really nice update..

[real.finder]

I being trying new update v3.1.2 i been facing some difficulties to make prefilter 4 works but after reading about it i found a i need an function call knlmeansCL for that now prefilter has gone threw but one more error poped up i m still struggeling to find out what is Mfilter and how it can be used if i want to . I tried Mfilter=true but seems like MFilter doesn't work like this i get this error "the name arguement "mfilter" to SmDgrain had the wrong type" .Please Dogway can u tell how MFilter could be used, thank you.
Really nice update..

mfilter must be clip

@Dogway is the next update will be delay more?

[Dogway]

Among other things I've been working on Stab2() and ResizersPack.
The most critical thing currently for v3.2d which is a fix, is prefilter=4, its temporal component isn't very good so in v3.2d it's motion compensated. Everything else are features so you don't have to worry. I hope to release about October after I finish a current project that's been dragging for more than I wished.

[RRAH]

Hello @Dogway

From the documentation:

Recommendation: Use slightly greater values than you are used to in other mdegrain based functions since the internal clip for the motion search in SMDegrain() has the contrast increased.

Can we say that thSAD=200 in other MVTools based functions corresponds with thSAD=300 in SMDegrain?

Thanks

[Dogway]

More like 1.25 times.

As a sidenote in the still unreleased 3.2d version I did some attempts at an automatic contrast prefilter for optimal motion search, this yielded to a thSAD 1.75 times increase for similar denoise results in that specific scene. The problem is that dynamic contrast implies dynamic SAD (based on source contrast instead of source noise/grain), so thSAD would not mean much as it's a fixed value in time. It's pity I might have to discard this concept because it was working very well on typical low contrast areas (clouds, details/lineart on shaded areas, etc).

[tormento]

Is there any plan to port it in VapourSynth?

[Dogway]

There's a port here. Not sure up to what version it covers up.

[tormento]

I have a problem with SMDegrain and AviSynth+, look here

[Dogway]

I can't guarantee support for AVS+ because I don't use it, but have you tried to run plain KNLMeansCL (no SMDegrain)?

[tormento]

I can't guarantee support for AVS+ because I don't use it, but have you tried to run plain KNLMeansCL (no SMDegrain)?

I have tried even with prefilter=2.
Plain KNLMeansCL(D=1, A=1, h=7.0,device_type="GPU") works ok.

[Dogway]

Try lsb_inout=true with Y8 format.

[tormento]

Try lsb_inout=true with Y8 format.

Dummies' version, i.e. please translate with example

[Dogway]

ConverttoY8()
Dither_convert_8_to_16()
KNLMeansCL(D=1, A=1, h=7.0,lsb_inout=true,device_type="GPU")
Ditherpost()

Try with other KNLMeansCL builds too.

[tormento]

ConverttoY8()
Dither_convert_8_to_16()
KNLMeansCL(D=1, A=1, h=7.0,lsb_inout=true,device_type="GPU")
Ditherpost()

Try with other KNLMeansCL builds too.

As I told you it's not a KNLMeansCL problem. The same effect appears even with prefilter=2.

[Dogway]

Then it's a AVS+ problem.

[tormento]

Then it's a AVS+ problem.

Tested on 64 bit. Same problem.

[tormento]

There's a port here. Not sure up to what version it covers up.

3.0d.

Do you know python? =)

[Dogway]

64 bit AVS+? Why don't you use plain avisynth?
I don't do programming, only some scripting languages.

[tormento]

64 bit AVS+? Why don't you use plain avisynth?
I don't do programming, only some scripting languages.

As there is no AviSynth 2.6.0 MT x64 and I have problems in windows 10 x64 to put together DGDecNV and KNLMeansCL in the same script on plain 32 bit. It leads to Cuda errors and green screen images.

Everything is perfect in x64 but AVS+ screen corruption on some scene change, almost black to white or something like that.

[Dogway]

You are trying to put together many sensible stuff, Windows 10, x64 avisynth, AVS+... You are paying the tax for using brand new interfaces. Wait until things iron out or deal with "early adopter" bugs.

In my opinion x64 never took off, I use MP_Pipeline for that and it works fine. Anyway, prefilters are things I do for convenience, but I can't really assume issues for thirds plugins. You should narrow the problem and ask the original developer. Prefilter=2 uses Quantile, so maybe AVS+ doesn't support it, little things like these are what probably refrained me from switching to AVS+.

[tormento]

You are trying to put together many sensible stuff, Windows 10, x64 avisynth, AVS+... You are paying the tax for using brand new interfaces. Wait until things iron out or deal with "early adopter" bugs.

The only new variable in the equation is Windows 10. AVS+ was introduced to get 64 bit environment for further testing between DGDecNV and KNLMeansCL.

In my opinion x64 never took off, I use MP_Pipeline for that and it works fine.

Can you please write some example of usage with SMDegrain?

Prefilter=2 uses Quantile, so maybe AVS+ doesn't support it, little things like these are what probably refrained me from switching to AVS+.

The image corruption happens with 2 AND 4. Other modes not tested.

[tormento]

I wrote this post about image corruption.

I think it could have same bonding.

[tormento]

I did a try with:

SMDegrain (tr=6,PreFilter=4,thSAD=600,contrasharp=false,refinemotion=false,lsb_out=true,plane=0,chroma=false,Globals=2)
SMDegrain (tr=6,            thSAD=600,contrasharp=false,refinemotion=false,lsb_in =true,plane=0,chroma=false,Globals=1,mode=6) 

and it hangs. It works with PreFilter=2.

Ideas?

[Dogway]

Test in avisynth with Win7.

[tormento]

Test in avisynth with Win7.

My reference machine is Windows 8.1. Windows 10 is only for testing one. No Windows 7 available =/

Can you please test it for me?

[Dogway]

Why would I test it for you?

[tormento]

Why would I test it for you?

I thought you had a Win7 system.

Anyhow, I think you release new version not only to get new feature but also to correct bugs. Am I wrong?

[Dogway]

Why don't you ask Khanattila? He developed KNLMeansCL.

[poisondeathray]

Tormento - it works for me on windows 8

What are your input clip dimensions and colorspace?

colorbars(pixel_type="yv12")
SMDegrain (tr=6,PreFilter=4,thSAD=600,contrasharp=false,refinemotion=false,lsb_out=true,plane=0,chroma=false,Globals=2)
SMDegrain (tr=6,            thSAD=600,contrasharp=false,refinemotion=false,lsb_in =true,plane=0,chroma=false,Globals=1,mode=6)

Not sure if it makes a difference, but I modified SMDegrain to use device_id =1 for KNLMeansCL by default (because on Intel CPU, the iCPU will usually be device_id = 0 and usually default openCL device because it's the first one)

In 3.1.2.d, it's line 182

                                              Y8      = Y8.KNLMeansCL(D=1, A=1, h=7.0,lsb_inout=true,device_type="GPU",device_id = 1)