vcfreq a plugin for vapoursynth

[vcmohan]

vcfreq is a plugin for vapoursynth. It operates in frequency domain to filter out certain types of regular noises. A large range of cascading filters can be specified.
The functions available in this plugin are

1.F1Quiver
2.F2Quiver

Full description of this plugin is at vcfreq
This plugin requires a fftw3f.3.dll to be present in path. Currently works in windows only.
Comments, suggestions are welcome

[feisty2]

is there some memory leak in the plugin?
the plugin consumed almost all 16GB of the RAM of my computer as the processing went on and then terminated by the OS...

[feisty2]

@vcmohan
I tried to fix the memory leak but...
can't even compile the untouched source code, there're like.. errors everywhere...
some tips to make it work plz?

[vcmohan]

Sorry for that. Is it that all functions in it have memory leak or did you try a particular one. If so can you give the script used?
Offhand I can see one possible problem. vcfreq assumes the number of cpus to be the number of threads active. This is communicated to the fftw3 dll.

[feisty2]

I've only used vcfreq.Sharp so far cuz I need to perform a deconvolution..
the script's just

core.vcfreq.Sharp (clip, line=0, wn=0.99, x=1, y=1, fr=25, scale=0.0059)

vspipe kept consuming RAM right from the very first frame and crashed around frame 2800 on a GrayS 720x576 vid as there was no more RAM left..

[vcmohan]

I have uploaded the corrected version of vcfreq plugin. Please check if it is working ok. These were compiled on 201m community visual studio compiler.

[feisty2]

the new version works!

[sl1pkn07]

any chance for can be builded on linux?

greetings

[Myrsloik]

Here's my modified source and x64 dll that should be compilable on linux if you have a float fftw3 lying around. I didn't test it though. Report your findings and so on. Just done as a 30 min code stabbing session while eating dinner.

Changes:
Removes the loadlibrary loading of fftw3 and instead uses the official fftw3.h
Replaced all homegrown pi constants with the more accurate M_PI
Removed all unused variables where it was obvious no use was ever intended
Fixed an extreme number of type conversion warnings, some of them by making sure all calculations stay in float precision

And can someone check this piece of code?

void pointNotchFilter2D( float * freqFilter, int* spec, int ht, int wd)
{
	// point centered notch filter
	float xf = spec[2] * wd / (NYQUIST);

	float yf = spec[3] * ht / (NYQUIST );

	int degree = 2 * spec[4];

	// calculate notch radial sqr
	float d0sq = sqrt (xf*xf + yf*yf);
	d0sq = (ht * ht + wd * wd)/ 4;

It looks very suspect. Like it should be multiplied or divided or something instead of the first value assigned to d0sq just being discarded.

[sl1pkn07]

with GCC:

In file included from vcfreq.cpp:66:0:
FQSharphelper.cpp: In function 'void getRealInput(float*, const finc*, int, int, int, int, int, int, bool)':
FQSharphelper.cpp:251:30: error: too many arguments to function 'int getSign(int)'
     data1[ w] = getSign( h, w) * fptr1[w];
                              ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In function 'void getRealOutput(float*, finc*, int, int, int, int, int, bool, finc, finc)':
FQSharphelper.cpp:319:39: error: too many arguments to function 'int getSign(int)'
     int val = data1[ w] * getSign(h, w);
                                       ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealInput(float*, const finc*, int, int, int, int, int, int, bool) [with finc = unsigned char]':
FQBlur.cpp:254:78:   required from here
FQSharphelper.cpp:251:24: error: too many arguments to function 'int getSign(int)'
     data1[ w] = getSign( h, w) * fptr1[w];
                        ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealOutput(float*, finc*, int, int, int, int, int, bool, finc, finc) [with finc = unsigned char]':
FQBlur.cpp:267:51:   required from here
FQSharphelper.cpp:319:34: error: too many arguments to function 'int getSign(int)'
     int val = data1[ w] * getSign(h, w);
                                  ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealInput(float*, const finc*, int, int, int, int, int, int, bool) [with finc = short unsigned int]':
FQBlur.cpp:278:80:   required from here
FQSharphelper.cpp:251:24: error: too many arguments to function 'int getSign(int)'
     data1[ w] = getSign( h, w) * fptr1[w];
                        ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealOutput(float*, finc*, int, int, int, int, int, bool, finc, finc) [with finc = short unsigned int]':
FQBlur.cpp:291:51:   required from here
FQSharphelper.cpp:319:34: error: too many arguments to function 'int getSign(int)'
     int val = data1[ w] * getSign(h, w);
                                  ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealInput(float*, const finc*, int, int, int, int, int, int, bool) [with finc = float]':
FQBlur.cpp:302:80:   required from here
FQSharphelper.cpp:251:24: error: too many arguments to function 'int getSign(int)'
     data1[ w] = getSign( h, w) * fptr1[w];
                        ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
FQSharphelper.cpp: In instantiation of 'void getRealOutput(float*, finc*, int, int, int, int, int, bool, finc, finc) [with finc = float]':
FQBlur.cpp:316:51:   required from here
FQSharphelper.cpp:319:34: error: too many arguments to function 'int getSign(int)'
     int val = data1[ w] * getSign(h, w);
                                  ^
FQSharphelper.cpp:217:5: note: declared here
 int getSign(int i)
     ^
make: *** [Makefile:2: all] Error 1

with clang

In file included from vcfreq.cpp:66:
./FQSharphelper.cpp:251:17: error: no matching function for call to 'getSign'
                                data1[ w] = getSign( h, w) * fptr1[w];
                                            ^~~~~~~
./FQSharphelper.cpp:217:5: note: candidate function not viable: requires single argument 'i', but 2 arguments were provided
int getSign(int i)
    ^
./FQSharphelper.cpp:319:27: error: no matching function for call to 'getSign'
                                int val = data1[ w] * getSign(h, w);
                                                      ^~~~~~~
./FQSharphelper.cpp:217:5: note: candidate function not viable: requires single argument 'i', but 2 arguments were provided
int getSign(int i)
    ^
2 errors generated.
make: *** [Makefile:2: all] Error 1

note: #include "FQSharpHelper.cpp" -> #include "FQSharphelper.cpp" in vcfreq.cpp (case sensitive)

[Myrsloik]

I think I've fixed that now. Redownload and try again. I think the other overload was declared too late or something.

[sl1pkn07]

done.

now build ok in linux ^o^

[vcmohan]

Originally Posted by Myrsloik

And can someone check this piece of code?

void pointNotchFilter2D( float * freqFilter, int* spec, int ht, int wd)
{
	// point centered notch filter
	float xf = spec[2] * wd / (NYQUIST);

	float yf = spec[3] * ht / (NYQUIST );

	int degree = 2 * spec[4];

	// calculate notch radial sqr
	float d0sq = sqrt (xf*xf + yf*yf);
	d0sq = (ht * ht + wd * wd)/ 4;

It looks very suspect. Like it should be multiplied or divided or something instead of the first value assigned to d0sq just being discarded.

Thanks for pointing out problematic code. Sorry when I uploaded the revised dll in May I missed to upload the updated source. I have done that now. Yes instead of the included lte file
the official fftw3 header can be used, but I found it to be so difficult to understand started using Fizik's lte version with suitable modifications.
As regards numerous warnings, yes I am guilty and lazy. Thankful for correcting.

[StainlessS]

yes I am guilty and lazy

As are we all, we are all here a consortium of individuals all at common purpose.
Sometimes we dont make mistakes, sometimes we do, but, every time we advance
and take from what we have learnt, thank you VC Mohan, you really are pretty good, when
not making mistakes, and that is what really counts

EDIT: Big Kiss vc, XXX.

EDIT: I picture myself as the one on the left, and you being the girly one on the right (with the blue hair, whats with that?)
I think your character is way too aggressive, I feel maybe that I'm being abused, please control yourself.

[KingLir]

Originally Posted by Myrsloik

Here's my modified source and x64 dll

Changes:
Removes the loadlibrary loading of fftw3 and instead uses the official fftw3.h
Replaced all homegrown pi constants with the more accurate M_PI
Removed all unused variables where it was obvious no use was ever intended
Fixed an extreme number of type conversion warnings, some of them by making sure all calculations stay in float precision

Thanks! This builds OK also on macOS with the following:

gcc -c -fPIC  ${CFLAGS} ${CPPFLAGS} -I. $(pkg-config --cflags vapoursynth) -o main.o vcfreq.cpp
g++ -dynamiclib -undefined suppress -flat_namespace *.o -o libvcfreq.dylib

[sl1pkn07]

the link is down @mirsloik

[Chthon]

Please kindly repost the fixed-for-linux source code.

[Are_]

https://www.dropbox.com/s/fyid7xne75....08.17.tar.bz2

[Chthon]

Thank you Are_ .

Edit: A few notes for anyone else trying to compile:

1. Looking at the newer source on the official page, it looks likes vc's fix for the issue pointed out in post #9 is to comment out both lines. This seems fine, since the variable is unused anyway.

2. It looks like there's another bug. The for loop at the bottom of Factorize.cpp is missing its {}s. I *think* the closing brace should go on line 52 (???).

3. libfftw3f_threads is a dependency (at least for the fixed version). If you don't link it at compile time, core.vcfreq.Sharp() will crash with an undefined symbol error for fftwf_init_threads.

4. I was able to compile with

g++ -Wall -shared -fPIC -I. -L/usr/local/lib/vsplugins/lib -lfftw3f_threads -o libvcfreq.so vcfreq.cpp

after copying the vapoursynth headers into the working directory.
(My libfftw3f_threads.so is in usr/local/lib/vsplugins/lib; adjust accordingly.)

[Myrsloik]

I updated my old and broken dropbox link. Not sure if I really want to fork the project since I'd have basically no time to maintain it anyway. Would be a shame if the changes were lost tough...