Thank you for your quick response.

Indeed, Paraview don't support 2D transfer function.

I will test that, thank you for give me the right file to edit.

Best regards,

Pierrick

2016-05-21 22:09 GMT+02:00 tom fogal < >:

Hi, && welcome,

On 05/21/2016 11:12 AM, Pierrick Pochelu wrote:

I am a french searcher in computing science and currently study
"transfer function" and specially "transfer function 2D". ImageVis3D is
the most powerful system I have found in this field. (Do you know other
system ? I know Paraview and Voreen too. )

Oh, I didn't know ParaView supported 2D transfer functions.

We're aware of Voreen of course; they have a slightly different focus. But we end up citing them and probably vice-versa (I don't pay much attention anymore).

Spatialized Transfer Function seems great. The idea is to color grey
points regarding their position in space. (x=red,y=green,z=blue). So we
keep spatial information in histogram which is currently lost. (Thus
maybe it is necessary to rise sampling in x and y axis.)

The idea is explained in this very interesting article :
http://www.cs.rug.nl/~roe/courses/wv/papers_final_assignments/Roettger2005.pdf

Figure 1 give an idea of what it would look like

*shrug* seems interesting.  I'm a bit wary, because even 2D transfer functions are sometimes a good idea and sometimes a bad idea; it depends on the dataset.  Certainly there are datasets that cannot be appropriately segmented via just value && gradient, though.

Were you asking how to implement it?  If you just wanted quick results, I would hack the shaders.  GLGridLeaper-Method-2D-L.glsl is where the 2D tfqn is sampled in the common case.  Looks like it already takes the volume's coordinates in model space, so you already have the input you need.
I'd be interested in how it turns out if you pursue it.

HTH,

-tom