Interestingly, though... if you bring up the radius, you would expect to get a DOG line generation, but you do not... instead it's almost as though the two images are being added? Not sure what's going on...
Patrick On Aug 16, 2010, at 2:47 AM, Patrick Sheffield wrote: > You, sir, are a genius... > > Thank you. It all makes sense now. Much much appreciated. > > Sincerely, > > Patrick > > On Aug 16, 2010, at 2:36 AM, Achim Breidenbach wrote: > >> Hi Patrick, >> >> the rendering "Sprite"-Patch within the GLSL-Patch needs to draw the quad in >> the correct dimensions. Those dimensions we only can get within the "Render >> in Image"-Patch with the "Image Dimensions"-Patch because the >> coordinate-system within a "Render In Image"-patch is different to the real >> worlds coordinate system. >> >> See the attached composition, which now shows all black because of your >> diff-patch ;-) >> >> <gaussian4.qtz> >> >> Best, >> >> Achim Breidenbach >> Boinx Software >> >> On 16.08.2010, at 11:23, Patrick Sheffield wrote: >> >>> Ahhhhhh.... now I'm getting it... so close, but for some reason, the image >>> is still being stretched - i.e. the aspect ratio is changing - see attached >>> - I put a difference blend patch to show what's happening: >>> >>> Patrick >>> >>> <gaussian3.qtz> >>> >>> On Aug 16, 2010, at 1:34 AM, Achim Breidenbach wrote: >>> >>>> Hi, >>>> >>>> there are some differences from texture2D to texture2DRect: in texture2D >>>> the xy-corrdinates can varying from 0.0 to 1.0, however in the >>>> texture2DRect you have to use pixel units. >>>> >>>> please find attached your composition with my suggestions. >>>> >>>> <gaussian2.qtz> >>>> >>>> best, >>>> >>>> Achim Breidenbach >>>> Boinx Software >>>> >>>> On 16.08.2010, at 10:17, Patrick Sheffield wrote: >>>> >>>>> Hmmmmm.... >>>>> >>>>> Well, I'm not sure how to do the solution you present in (1)... do I >>>>> multiply the gl_TexCoord[0].xy by my coordinates to correct? And when I >>>>> changed the sampler2D to sampler2DRect (and the textureSize2DRect, and >>>>> the texture2DRect...), I then got a uniform reddish gray image in my >>>>> viewer. >>>>> >>>>> Thanks for your continued help, I feel I am quite close... >>>>> >>>>> Patrick >>>>> >>>>> On Aug 16, 2010, at 12:48 AM, Achim Breidenbach wrote: >>>>> >>>>>> Hi Patrick, >>>>>> >>>>>> I just improved the composition from Miroslaw in pretty much the same >>>>>> way as your solution for the "gaussian" version of the filter looks >>>>>> like. Maybe you can fiddle in there your code and get rid of all your >>>>>> questions ;-) >>>>>> >>>>>> 1) It is, because you don't respect the gl_TextureMatrix of the the src >>>>>> texture. Textures can be flip, but the matrix always put them back >>>>>> correctly. You have to use the gl_TexCoord[0].xy in the fragment shader >>>>>> to get correct xy-coordinates, they are transformed in the Vertex Shader >>>>>> by the texture matrix. >>>>>> >>>>>> 2) the "default" implementation of the GLSL shader puts in a "uniform >>>>>> sampler2D texture;" which is misleading! This means, that the input for >>>>>> the texture has to be a square. so QC is converting your input image to >>>>>> a square texture and feeds this into the GLSL-shader (and manipulates >>>>>> the texture-matrix that way, that the image-aspect ratio is still >>>>>> correct!). Your code ready the native texture size with >>>>>> textureSize2D(src, 0) which is 512x512 I guess in your case. Replace all >>>>>> "sample2D" with "sample2DRect" to get your original texture in the >>>>>> correct size. >>>>>> >>>>>> 3) You have to use a "Render in Image"-Patch to render the GLSL-Shader >>>>>> into a new texture to use it in following patches. >>>>>> >>>>>> Best, >>>>>> >>>>>> Achim >>>>>> >>>>>> <compound-gaussian 1.qtz> >>>>>> >>>>>> On 16.08.2010, at 09:27, Patrick Sheffield wrote: >>>>>> >>>>>>> Thank you, Miroslaw - that works, I'd never played with an OpenCL >>>>>>> patch... >>>>>>> >>>>>>> Toward that end, I've been trying out some GLSL routines, but I have a >>>>>>> few questions for anyone out there: >>>>>>> >>>>>>> <gaussian1.qtz> >>>>>>> >>>>>>> First, why does this composition flip the image? >>>>>>> Why does it mess with the aspect ratio? >>>>>>> And how would I feed the output of a GLSL patch to do further >>>>>>> processing? >>>>>>> >>>>>>> I realize a Gaussian blur needn't be done in a GLSL patch - could be >>>>>>> handled in a CI kernel, etc... but I have other GLSL routines I'd like >>>>>>> to use. >>>>>>> >>>>>>> Thanks for any insight you might have, >>>>>>> >>>>>>> Patrick >>>>>>> >>>>>>> On Aug 15, 2010, at 5:31 AM, Miroslaw Rusin wrote: >>>>>>> >>>>>>>> thanks for the info. so it's something more like this then... >>>>>>>> >>>>>>>> <compound-3.qtz> >>>>>>>> >>>>>>>> >>>>>>>> it's slow - O(x*y*r²), where: >>>>>>>> x - image with >>>>>>>> y - image height >>>>>>>> r - radius >>>>>>>> >>>>>>>> can be O(x*y) with sum table >>>>>>>> >>>>>>>> M >>>>>>>> >>>>>>> >>>>>>> _______________________________________________ >>>>>>> Do not post admin requests to the list. They will be ignored. >>>>>>> Quartzcomposer-dev mailing list >>>>>>> ([email protected]) >>>>>>> Help/Unsubscribe/Update your Subscription: >>>>>>> http://lists.apple.com/mailman/options/quartzcomposer-dev/achim%40boinx.com >>>>>>> >>>>>>> This email sent to [email protected] >>>>>> >>>>> >>>> >>> >> > > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Quartzcomposer-dev mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > http://lists.apple.com/mailman/options/quartzcomposer-dev/psheffield%40earthlink.net > > This email sent to [email protected]
_______________________________________________ Do not post admin requests to the list. They will be ignored. Quartzcomposer-dev mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/quartzcomposer-dev/archive%40mail-archive.com This email sent to [email protected]
