I'm fairly new to this world and I'm currently looking for the right direction to proceed in for my project. Any suggestions would be very appreciated.

There will be x,y values of a single pixel coming in and I would like this to translate to some sort of 2D gaussian window say 25x25 for an alpha channel mask (i'm terrible at explaining this).. I have seen something on the list that I think leads in the right direction: 

https://cycling74.com/forums/index.php?t=msg&goto=127983&rid=0&srch=reveal+jitter#msg_127983

This uses 2 videos and has jit.lcd scrub away the top layer to reveal the bottom layer. Effectively, I would like the scrubbing process to have a gaussian window that can be cumulative. I.e. the first scrub only reveals some percentage of the alpha channel rather than a binary mask. and the more scrubbing will result in more being revealed... I am not so sure then how to do the cumulative part but I imagine at least getting the gaussian window part of it should be more straightforward.

But I am not so sure how the most efficient way of taking the pixel values from the jit.lcd for example and creating a window of values centered around that pixel. I would think a convolution would be nice, but it is very expensive to compute I believe, especially if i am asking for a 25x25 window. Also, I would eventually like this window to be cumulative. Is there something more clever? Perhaps with the gauss6x shaders, but then I would not be so keen if that would have to involve a readback?

Would all of these computations be better suited on a pixel shader? I'm slowly trying to learn glsl...

Any suggestions on how to proceed would be very grateful. Thanks!

I'm fairly new to this world and I'm currently looking for the right direction to proceed in for my project.  Any suggestions would be very appreciated. 

There will be x,y values of a single pixel coming in and I would like this to translate to some sort of 2D gaussian window say 25x25 for an alpha channel mask (i'm terrible at explaining this).. I have seen something on the list that I think leads in the right direction: https://cycling74.com/forums/index.php?t=msg&goto=127983&rid=0&srch=reveal+jitter#msg_127983 

This uses 2 videos and has jit.lcd scrub away the top layer to reveal the bottom layer.  Effectively, I would like the scrubbing process to have a gaussian window that can be cumulative.  I.e. the first scrub only reveals some percentage of the alpha channel rather than a binary mask.  and the more scrubbing will result in more being revealed... I am not so sure then how to do the cumulative part but I imagine at least getting the gaussian window part of it should be more straightforward.

But I am not so sure how the most efficient way of taking the pixel values from the jit.lcd for example and creating a window of values centered around that pixel.  I would think a convolution would be nice, but it is very expensive to compute I believe, especially if i am asking for a 25x25 window.  Also, I would eventually like this window to be cumulative. Is there something more clever?  Perhaps with the gauss6x shaders, but then I would not be so keen if that would have to involve a readback?

Would all of these computations be better suited on a pixel shader?  I'm slowly trying to learn glsl... 

Any suggestions on how to proceed would be very grateful.  Thanks!
pkmital.


alpha-channel-mask-for-video-with-cumulative-2d-gaussian-window

Got something working.. not too sure about how to accumulate values for the alpha channel though.

This page is really helpful: http://abstrakt.vade.info/?p=147


Actually just noticed that it is alpha-blending the wrong texture. Really strange.

Actually just noticed that it is alpha-blending the wrong texture.  Really strange.


The problem was my mis-understanding of what co.alphablend.jxs does. I was thinking the param amount 1. 1. 1. 1. would blend each rgba channel of the 2 textures based on that weight. So for my purposes, I just wanted a gpu version of a jit.unpack and jit.pack for the alpha channel. Changing the shader, setting the layers properly for each videoplane, and setting blend_enable 1 finally lead to something that made sense.

    vec4 a = texture2DRect(tex0, texcoord0);

    vec4 b = texture2DRect(tex1, texcoord1);

The problem was my mis-understanding of what co.alphablend.jxs does.  I was thinking the param amount 1. 1. 1. 1. would blend each rgba channel of the 2 textures based on that weight.  So for my purposes, I just wanted a gpu version of a jit.unpack and jit.pack for the alpha channel.  Changing the shader, setting the layers properly for each videoplane, and setting blend_enable 1 finally lead to something that made sense.

{   
	vec4 a = texture2DRect(tex0, texcoord0);
	vec4 b = texture2DRect(tex1, texcoord1);

alpha channel mask for video with cumulative 2d gaussian window?