Frfun7

Frfun7 is a spatial fractal denoising plugin by Copyright (C) 2002-2006, 2013 by Marc Fauconneau (prunedtree), (C)2021 Ferenc Pintér Additional work and fixes (C)2022 by dubhater.

Usage

frfun7 (clip, float "lambda", float "T", float "Tuv", int "P", int "TP", int "R1")

     clip   =

        Input clip. 


    float  lambda = 1.1

        Adjust the power of the local denoising.
        
        It must not be negative.


    float  T = 6.0

        Limits the max luma denoising power for edges; 0 disables processing. 
        
        It must not be negative.


    float  Tuv = 2.0

        Limits the max chroma denoising power for edges; 0 disables processing. 
        
        It must not be negative.


    int  P = 0

        By testing, one can conclude it's a "speed -vs- quality" trade off setting. 

                0 : faster but slightly lower quality than frfun7_rev6 (may create minor artifacts around line edges).
                1 : adaptive overlapping (see also TP1). slower than frfun7_rev6 but the quality is a little bit better.
                2 : temporal
                4 : adaptive radius

        Internally the parameter is treated as a bit mask but probably it has no point.

        Parameter is available since frfun7 2013.


    int  TP1 = 0

        A threshold which affects P=1 (adaptive overlapping).
        0 will always run into a final filtering part.
        Values greater than 0 will make it skip processing some pixels.
        (Introduced as a separate parameter in r0.7test. This value had to be encoded into P as TP1*1000 previously.)

    int  R1 = 3

        Radius for first pass of the internal algorithm.
        First pass in pre v0.7 was fixed to 3 (and was no separate parameter)
        Valid values are 2 or 3. 2 is faster.

frfun7 with default settings:

AviSource("Blah.avi")
frfun7(lambda=1.1, T=6.0, Tuv=2.0, P=0, TP1=0, R1=3)

Links

http://avisynth.nl/index.php/Frfun7

VapourSynth port:

https://github.com/dubhater/vapoursynth-frfun7

Other

Special thanks to Reel.Deel for the testing and comparison with previous versions and maintaining the documentation at avisynth.nl.

Build instructions

Visual Studio 2019:

use IDE

Windows GCC

(mingw installed by msys2) From the 'build' folder under project root:

del ..\CMakeCache.txt
cmake .. -G "MinGW Makefiles" -DENABLE_INTEL_SIMD:bool=on
cmake --build . --config Release 

Linux

from the 'build' folder under project root: ENABLE_INTEL_SIMD is automatically off for non-x86 architectures Note: plugin source supports non-INTEL compilation since v0.8

• Clone repo and build

    git clone https://github.com/pinterf/Frfun7
    cd Frfun7
    cmake -B build -S .
    cmake --build build
  

  Useful hints:
  build after clean:

  cmake --build build --clean-first
  

  Force no assembler support (valid since v0.8)

  cmake -B build -S . -DENABLE_INTEL_SIMD:bool=off
  

delete CMake cache

rm build/CMakeCache.txt

• Find binaries at

    build/Frfun7/libfrfun7.so
  

• Install binaries

    cd build
    sudo make install
  

History

Version         Date            Changes
0.9             2022/06/29      - Performance increase in Intel sse2 - backport from VapourSynth (dubhater):
                                  p=0 184 fps to 304 fps 65% faster
                                  p=1  36 fps to  61 fps 69% faster
                                  p=2 233 fps to 282 fps 21% faster
                                  p=4 263 fps to 401 fps 52% faster
                                  note: The output is slightly different at the left and right edges of the frame.
                                - additional fix bug when p=1 and tp1>0
0.8             2022/05/18      - backport some fixes and non-Intel C code from VapourSynth port by dubhater:
                                - Fix P=2 (temporal) blocky rectangular artifacts at the most top and bottom area
                                - Fix bug when p=1 and tp1>0
                                - Avoid loss of precision when p=1 or p=2, Fixes https://github.com/pinterf/Frfun7/issues/1
                                - Add C-only (other than Intel SIMD) code path, enable usage of ENABLE_INTEL_SIMD build option.

0.7             2021/09/09      - release

0.7 WIP         2021/05/25      - re-enable T=0, Tuv=0 cases (unprocessed plane copy)
                                - add experimental TP1 (default 0) a threshold for P=1 (temporal overlapping) mode
                                - add experimental R1 (default 3, can be set to 2) first pass radius

0.7 WIP         2021/05/20      - Source is based on a 2006/05/11 snapshot 
                                - Code refresh and additions by pinterf
                                - move to git: https://github.com/pinterf/Frfun7
                                - Source to VS2019 solution
                                - Add readme, usage based on Avisynth wiki
                                - Add/guess missing source parts, rename files
                                - Update AviSynth headers
                                - Add version resource
                                - Avisynth V2.6 style plugin
                                - Implement all mmx inline assembler as SIMD intrinsics
                                - x64 build
                                - fix some rounding and other issue

2013            2013/09/04      - no longer buffers the input; yields a nice speed increase.
                                - "P" parameter added
                                  Note: frfun7 2013 is the updated version, unfortunately the output is 
                                  not completely identical to frfun7_rev6, for that reason both versions
                                  are available. Read description of the P parameter for more information. 

rev6            2006/05/10      - bug fixes
                                - remove mod8 restriction
                                - process first and last frame

rev1            2006/05/05      - initial release.