Re: [Gegl-developer] Gegl API calls vs. Gimp UI - different filter results

[Florian Klemme <mail florianklemme de>]

Just to have this one closed: I've found the reasons of my problems.
a) Height difference: This is actually intended by Gegl's area filter 
class. Just overload get_bounding_box an make it pass through the source 
bounding box to avoid any increases of your result rectangle size.
b) Result difference: In this case, I had to use babl's R'G'B' instead 
of RGB.

_Florian

Am 24.11.2014 um 17:20 schrieb Florian Klemme:
> Hi,
>
> I am just starting out building some Gegl filters and I'm experiencing 
> some problems with different results, depending on from where I invoke 
> the filters. I'm using Ubuntu 14.04, Gegl from git (tagged 0.2.0), 
> current babl from git (and Gimp 2.8 from the Ubuntu repository, just 
> for testing purposes).
>
> To test the results of my filters, I wrote a very simple program, 
> which basically looks like this:
>
>             GeglNode *root = gegl_node_new();
>             GeglNode *load = gegl_node_new_child(root,
>                 "operation", "gegl:load",
>                 "path",      "test_input.png",
>                 NULL);
>             GeglNode *test = gegl_node_new_child(root,
>                 "operation", test_operations[i],
>                 NULL);
>             GeglNode *save = gegl_node_new_child(root,
>                 "operation", "gegl:save",
>                 "path",      output_filename,
>                 NULL);
>             gegl_node_link_many(load, test, save, NULL);
>             gegl_node_process(save);
>
> My first problem is that if I use this code, my output file resolution 
> is increasing. When I run my filter and print out some debug 
> information, I can see that the extra amount of pixels I request in my 
> prepare function is already applied to the value of the result 
> argument in my process function. The call to 
> gegl_operation_get_required_for_output returns me an even larger 
> rectangle.
>
> This is a brief excerpt from my gegl filter code:
>
> static void prepare(GeglOperation *operation) {
>     GeglOperationAreaFilter *area = 
> GEGL_OPERATION_AREA_FILTER(operation);
>     GeglChantO              *o    = GEGL_CHANT_PROPERTIES(operation);
>
>     // Extra pixels needed in each direction
>     area->left = area->right = 0;
>     area->top = area->bottom = o->height; // sloppy, but at least enough
>
>     gegl_operation_set_format(operation, "input", babl_format("RGB 
> float"));
>     gegl_operation_set_format(operation, "output", babl_format("RGB 
> float"));
> }
>
> process (...
>          const GeglRectangle *result,
>          ...)
> {
>     const GeglChantO *o = GEGL_CHANT_PROPERTIES(operation);
>
>     GeglRectangle source =
>         gegl_operation_get_required_for_output(operation, "input", 
> result);
>
>     g_print("result - x: %i, y: %i, width: %i, height: %i\n",
>         result->x, result->y, result->width, result->height);
>     g_print("source - x: %i, y: %i, width: %i, height: %i\n",
>         source.x, source.y, source.width, source.height);
>     ...
>
> With the same input file (640 x 480), the same parameters and on the 
> same machine...
> A) Run with the program shown above:
>     result - x: 0, y: -10, width: 256, height: 500
>     source - x: 0, y: -20, width: 256, height: 520
>     result - x: 256, y: -10, width: 384, height: 500
>     source - x: 256, y: -20, width: 384, height: 520
> B) Run within Gimp via UI
>     result - x: 0, y: 0, width: 256, height: 480
>     source - x: 0, y: -10, width: 256, height: 500
>     result - x: 256, y: 0, width: 384, height: 480
>     source - x: 256, y: -10, width: 384, height: 500
>
> Run with the program above, the output image will have an additional 
> height of two times o->height (which is in this test cases always 10). 
> The result are two (black) bars above and below the actual image. This 
> problem does *not* occur when I run this filter within Gimp!
>
> My second problem - but I don't care very much about this at the 
> moment - is that the colors of the result are not quite the same as 
> well, although I'm letting babl do all that work (I think). Because of 
> the effect of the filter, the difference in brightness is quite easy 
> to see.
>
> I hope you have any idea that could help avoid the increase of *result 
> / the output image in the first place. I appreciate every hint!
>
> Regards,
> Florian
> _______________________________________________
> gegl-developer-list mailing list
> List address:    gegl-developer-list gnome org
> List membership: 
> https://mail.gnome.org/mailman/listinfo/gegl-developer-list